CN118075427A - Intelligent monitoring management method and system - Google Patents

Abstract

The present invention proposes an intelligent monitoring management method and system. Belonging to the field of monitoring management technology, the method includes: training a monitoring model in a monitoring scenario through a deep learning algorithm, the monitoring model includes a target recognition model and a behavior analysis model, and deploying the trained monitoring model on a front-end intelligent monitoring device; the monitoring device inputs the collected first monitoring data into the monitoring model, and uses the monitoring model to extract targets and judge behaviors of the collected first monitoring data in real time, obtains a first analysis result, and sends the first analysis result and key frames to an edge computing node. By training the monitoring model through a deep learning algorithm, intelligent processing of monitoring data can be achieved, which can greatly reduce the need for manual intervention, improve monitoring efficiency, and reduce the cost and error rate of manual monitoring.

Description

Intelligent monitoring management method and system

Technical Field

The invention proposes an intelligent monitoring management method and system, belonging to the technical field of monitoring management.

Background technique

With the rapid development of information technology and the popularization of intelligent applications, monitoring management has become an important part of the modern social security system. Traditional monitoring management methods rely on manual viewing and analysis of monitoring images, which is not only inefficient, but also prone to omissions of safety hazards due to human fatigue and negligence. In addition, with the rapid increase in the number of monitoring devices, the massive amount of monitoring data has also brought great challenges to data storage and analysis.

Therefore, how to effectively process and analyze monitoring data, improve monitoring efficiency, and reduce safety hazards is an urgent problem to be solved in the current monitoring management field.

Summary of the invention

The present invention provides an intelligent monitoring management method and system to solve the problems mentioned in the above background technology:

The present invention provides an intelligent monitoring and management method, the method comprising:

The monitoring model in the monitoring scenario is trained by a deep learning algorithm, wherein the monitoring model includes a target recognition model and a behavior analysis model, and the trained monitoring model is deployed on a front-end intelligent monitoring device;

The monitoring device inputs the collected first monitoring data into the monitoring model, performs target extraction and behavior judgment on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and the key frame to the edge computing node;

The edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules to obtain second monitoring data, and uploads the second monitoring data to the cloud server;

The cloud server stores the received second monitoring data, further analyzes and mines the second monitoring data by performing a big data analysis algorithm, generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel.

Furthermore, the monitoring model in the monitoring scenario is trained by a deep learning algorithm, the monitoring model includes a target recognition model and a behavior analysis model, and the trained monitoring model is deployed on a front-end intelligent monitoring device, including:

Obtaining public surveillance video clips containing multiple target types and behavior patterns from the Internet through an API interface, annotating the surveillance clips, obtaining target recognition and behavior analysis data sets, and using 80% of the data sets as the first data set and 20% as the second data set;

Inputting the labeled first data set into a deep neural network architecture, and training an object recognition model through the deep neural network architecture;

Constructing a behavior analysis model suitable for behavior analysis, and training the behavior analysis model using a first data set;

The target recognition model and the behavior analysis model are integrated and optimized, and some underlying features of the target recognition model and the behavior analysis model are shared through joint training and multi-task learning, and their respective results are outputted;

Evaluate the performance indicators of the trained monitoring model using the second data set, and adjust and optimize the model according to the evaluation results;

The adjusted and optimized monitoring model is converted and deployed in the front-end intelligent monitoring device.

Further, the monitoring device inputs the collected first monitoring data into the monitoring model, performs target extraction and behavior judgment on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and the key frame to the edge computing node, including:

The front-end intelligent monitoring device collects the video stream through the built-in camera, uses the collected video stream as the first monitoring data, pre-processes the first monitoring data, and divides the continuous video stream into single-frame images;

Each frame of the image is input into the deployed target recognition model, which performs target detection in real time based on the trained parameters and outputs target detection information. The main task of this stage is to accurately identify and extract the target object of focus from the complex background.

Input a continuous multi-frame image sequence into the behavior analysis model, and determine whether the target has abnormal behavior or specific behavior patterns by analyzing the target object's motion trajectory and morphological changes in the time dimension;

According to the results of target detection and behavior analysis, a first analysis result is generated by summarizing, and frames where behavior events occur or frames with key information are selected and saved as key frames;

The first analysis result and the selected key frame are encrypted and compressed, and the encrypted and compressed first analysis result and the selected key frame are transmitted to the edge computing node through a multi-channel transmission protocol.

Furthermore, the edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules to obtain second monitoring data, and uploads the second monitoring data to the cloud server; the second monitoring data includes abnormal events or important information, including:

The edge computing node receives the first analysis result and key frame data sent from the front-end intelligent monitoring device, and performs decryption and decompression;

According to preset rules, the target information and behavior events in the first analysis result are preliminarily screened to eliminate false positives and data that do not meet the conditions, and verify whether the content of the key frame matches the analysis result;

The filtered data is analyzed and optimized again through a parallel processing algorithm, and according to preset business rules and security policies, abnormal events are identified and marked from the optimized data, and information on key time and space points is extracted to obtain the second monitoring data.

If there are multiple monitoring sources, the edge computing node is used to fuse and correlate cross-source data to integrate related events or information from different sources;

The second monitoring data is compressed and encrypted, and the packaged second monitoring data is prioritized according to the severity of the abnormal event or the urgency level of the important information; and the compressed and encrypted second monitoring data is transmitted to the cloud server according to the priority through a multi-channel transmission protocol.

Furthermore, the cloud server stores the received second monitoring data, further analyzes and mines the second monitoring data by performing a big data analysis algorithm, generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel, including:

The cloud server receives the second monitoring data uploaded by the edge computing node, decrypts and decompresses the second monitoring data, and stores the second monitoring data in different storage spaces according to the priority order;

The cloud server preprocesses the second monitoring data stored in different storage spaces again, and performs feature engineering processing on the preprocessed second monitoring data;

The pre-processed data is analyzed in depth through a machine learning algorithm to obtain a second analysis result; and a third analysis result is obtained based on spatiotemporal sequence analysis, cluster analysis, and association rule analysis algorithms;

If it is a specific scenario, a targeted analysis is performed through domain knowledge and an expert system to obtain a fourth analysis result;

Based on various analysis results, a comprehensive situation model is constructed, and the regional situation of the comprehensive situation model is evaluated to obtain the fourth analysis result;

The first to fourth analysis results are summarized and a comprehensive situation report is generated. The comprehensive situation report is pushed to relevant managers in various ways. After receiving the comprehensive situation report, the relevant managers perform corresponding processing.

The present invention proposes an intelligent monitoring and management system, the system comprising:

Model training module: trains the monitoring model in the monitoring scenario through deep learning algorithms, the monitoring model includes a target recognition model and a behavior analysis model, and deploys the trained monitoring model on the front-end intelligent monitoring device;

Data transmission module: The monitoring device inputs the collected first monitoring data into the monitoring model, performs target extraction and behavior judgment on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and key frames to the edge computing node;

Data upload module: the edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules, obtains second monitoring data, and uploads the second monitoring data to the cloud server;

Result feedback module: The cloud server stores the received second monitoring data, and further analyzes and mines the second monitoring data by performing a big data analysis algorithm, and generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel.

Furthermore, the model training module includes:

Dataset division module: obtain public surveillance video clips containing multiple target types and behavior patterns from the Internet through the API interface, annotate the surveillance clips, obtain target recognition and behavior analysis data sets, and use 80% of the data sets as the first data set and 20% as the second data set;

Data input module: inputting the annotated first data set into the deep neural network architecture to train the target recognition model through the deep neural network architecture;

Model building module: building a behavior analysis model suitable for behavior analysis, and training the behavior analysis model through the first data set;

Fusion optimization module: Fusion and optimization of the target recognition model and the behavior analysis model. Through joint training and multi-task learning, some underlying features of the target recognition model and the behavior analysis model are shared and their respective results are output.

Performance evaluation module: evaluates the performance indicators of the trained monitoring model through the second data set, and adjusts and optimizes the model according to the evaluation results;

Model conversion module: converts the adjusted and optimized monitoring model and deploys the converted monitoring model in the front-end intelligent monitoring device.

Furthermore, the data transmission module includes:

Data acquisition module: The front-end intelligent monitoring device acquires the video stream through the built-in camera, uses the acquired video stream as the first monitoring data, pre-processes the first monitoring data, and divides the continuous video stream into single-frame images;

Target detection module: Each frame of the image is input into the deployed target recognition model, which performs target detection in real time based on the trained parameters and outputs target detection information. The main task of this stage is to accurately identify and extract the target object of focus from the complex background.

Behavior judgment module: inputs a continuous multi-frame image sequence into the behavior analysis model, and judges whether the target has abnormal behavior or specific behavior patterns by analyzing the movement trajectory and morphological changes of the target object in the time dimension;

Saving module: according to the results of target detection and behavior analysis, the first analysis result is generated by summarizing, and the frames where the behavior events occur or have key information are selected and saved as key frames;

Encryption and compression module: encrypt and compress the first analysis result and the selected key frame, and transmit the encrypted and compressed first analysis result and the selected key frame to the edge computing node through a multi-channel transmission protocol.

Furthermore, the data uploading module includes:

Decryption and decompression module: The edge computing node receives the first analysis result and key frame data sent by the front-end intelligent monitoring device, and performs decryption and decompression;

Preliminary screening module: Preliminary screening of target information and behavior events in the first analysis result according to preset rules, eliminating false positives and data that do not meet the conditions, and verifying whether the content of the key frame matches the analysis result;

Secondary optimization module: The filtered data is analyzed and optimized again through a parallel processing algorithm, and according to the preset business rules and security policies, abnormal events are identified and marked from the optimized data, and information on key time and space points is extracted to obtain the second monitoring data.

Integration module: If there are multiple monitoring sources, the edge computing node is used to fuse and correlate cross-source data, and integrate related events or information from different sources;

Priority sorting module: compress and encrypt the second monitoring data, prioritize the packaged second monitoring data according to the severity of the abnormal event or the urgency level of important information; and transmit the compressed and encrypted second monitoring data to the cloud server according to priority through a multi-channel transmission protocol.

Furthermore, the result feedback module includes:

Space allocation module: the cloud server receives the second monitoring data uploaded by the edge computing node, decrypts and decompresses the second monitoring data, and stores the second monitoring data in different storage spaces according to the priority order;

Preprocessing module: the cloud server re-preprocesses the second monitoring data stored in different storage spaces; and performs feature engineering processing on the re-preprocessed second monitoring data;

Analysis result module: The pre-processed data is analyzed in depth through machine learning algorithms to obtain the second analysis result; and the third analysis result is obtained based on spatiotemporal sequence analysis, cluster analysis and association rule analysis algorithms;

The third analysis module: if it is a specific scenario, targeted analysis is performed through domain knowledge and expert system, and the fourth analysis result is obtained;

Fourth analysis module: Based on various analysis results, a comprehensive situation model is constructed, and the regional situation of the comprehensive situation model is evaluated to obtain the fourth analysis result;

Report feedback module: summarize the first to fourth analysis results and generate a comprehensive situation report, which is pushed to relevant managers in various ways; relevant managers take corresponding actions after receiving the comprehensive situation report.

Beneficial effects of the present invention: by training the monitoring model through the deep learning algorithm, it is possible to realize the intelligent processing of monitoring data, which can greatly reduce the need for manual intervention, improve monitoring efficiency, and reduce the cost and error rate of manual monitoring; the monitoring model is deployed on the front-end intelligent monitoring equipment, and the collected monitoring data can be extracted and the behavior judged in real time; by introducing the edge computing node, the secondary screening and optimization of the first analysis results and key frames are realized; the amount of data transmission can be reduced, the network burden can be reduced, and valuable information can be further screened out, and the data processing efficiency can be improved; the second monitoring data is stored through the cloud server, and further analyzed and mined through the big data analysis algorithm; the long-term storage and backup of data can be realized, and potential safety hazards and trends can be discovered through big data analysis, providing strong support for decision-making; the generated comprehensive situation report can fully reflect the real-time situation and potential risks of the monitoring scene, and provide decision-making basis for relevant personnel. This comprehensive situation perception and feedback mechanism helps to improve the level of security prevention and reduce the occurrence of security incidents.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a diagram showing steps of the method of the present invention;

FIG. 2 is a system module diagram of the present invention.

Detailed ways

In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

In the following description, many specific details are set forth to facilitate a full understanding of the present invention. The embodiments described are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

One embodiment of the present invention, as shown in FIG1 , is an intelligent monitoring management method, the method comprising:

S1. Train the monitoring model in the monitoring scenario through a deep learning algorithm, wherein the monitoring model includes a target recognition model and a behavior analysis model, and deploy the trained monitoring model on a front-end intelligent monitoring device;

S2. The monitoring device inputs the collected first monitoring data into the monitoring model, performs target extraction and behavior judgment on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and the key frame to the edge computing node;

S3, the edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules to obtain second monitoring data, and uploads the second monitoring data to the cloud server; the second monitoring data includes abnormal events or important information;

S4. The cloud server stores the received second monitoring data, further analyzes and mines the second monitoring data by performing a big data analysis algorithm, generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel.

The working principle of the above technical solution is: use the deep learning algorithm to train the data set in the monitoring scene and establish a monitoring model, which includes a target recognition model and a behavior analysis model. The target recognition model is used to accurately identify various targets (such as people, vehicles, etc.) from the video stream, and the behavior analysis model is responsible for analyzing the behavioral characteristics and activity patterns of the target; the trained monitoring model is deployed on the front-end intelligent monitoring device. The monitoring device captures the monitoring screen in real time as the first monitoring data and inputs it into the monitoring model for real-time analysis. The monitoring model performs target extraction in real time, identifies and locates each target in the monitoring screen, and determines whether the target's behavior has abnormalities or important features based on the behavior analysis model to form a preliminary first analysis result. At the same time, the key frame (i.e., the frame containing important information) will be extracted and sent to the edge computing node together with the first analysis result. The edge computing node performs secondary screening and optimization of the first analysis results and key frames transmitted from the front end according to the preset rules, eliminates redundant information, enhances the quality of effective information, and then generates more refined and targeted second monitoring data. The second monitoring data mainly focuses on abnormal events or important information in the monitoring scene, reduces the amount of original data transmission, and improves data processing efficiency. After receiving the second monitoring data uploaded by the edge computing node, the cloud server will store and manage it for a long time and on a large scale; it will use big data analysis algorithms to deeply mine and analyze these data to find potential correlations, patterns and trends, thereby forming a comprehensive situation report covering the overall monitoring situation. Finally, the cloud server will feed back the generated comprehensive situation report to relevant personnel through appropriate channels for decision-making reference, emergency response or daily management.

The effects of the above technical solution are: the target recognition model and behavior analysis model trained by the deep learning algorithm can achieve high-precision target detection and behavior understanding, improve the intelligence level of the monitoring system, and accurately capture the detailed information in the monitoring scene; the front-end intelligent monitoring equipment is equipped with a trained model, which can realize real-time target extraction and behavior judgment, quickly discover potential safety hazards or abnormal situations, and help improve the response speed to emergencies. By introducing edge computing nodes, a large amount of monitoring data generated by the front-end equipment can be processed locally and quickly, and truly valuable data (second monitoring data) can be screened out, and invalid data transmission can be reduced. At the same time, the network bandwidth pressure is reduced, and the timeliness and accuracy of data processing are improved. The second monitoring data pre-processed by the edge computing node is uploaded to the cloud server, so that the cloud platform does not need to process all the original data, but focuses on in-depth analysis of key abnormal events or important information, which improves the effective use of cloud computing resources. The cloud server uses big data analysis technology to deeply mine the second monitoring data, which can reveal the patterns and trends hidden in the massive monitoring information, form a comprehensive situation report, and help managers fully grasp the overall security situation and activity rules of the monitoring area. Comprehensive situation reports can provide powerful decision-making support for relevant departments and personnel, promote rapid and scientific decision-making, and take timely and effective actions to deal with various security issues or emergencies.

In one embodiment of the present invention, the monitoring model in the monitoring scenario is trained by a deep learning algorithm, the monitoring model includes a target recognition model and a behavior analysis model, and the trained monitoring model is deployed on a front-end intelligent monitoring device, including:

Obtain public surveillance video clips containing multiple target types and behavior patterns from the Internet through an API interface, annotate the surveillance clips, obtain target recognition and behavior analysis data sets, and use 80% of the data sets as the first data set and 20% as the second data set; the data sets cover but are not limited to different lighting conditions, weather conditions, crowd density, and target size and motion changes;

The labeled first data set is input into a deep neural network architecture (e.g., YOLO, Faster R-CNN), and the target recognition model is trained by the deep neural network architecture; the trained target recognition model can accurately locate and classify different target objects, such as people, vehicles, specific markers, etc., from the original video frames.

Constructing a behavior analysis model suitable for behavior analysis (such as a spatiotemporal convolutional network (STCNN), a recurrent neural network (RNN), or a long short-term memory network (LSTM)), wherein the behavior analysis model is used to capture and understand motion trajectories and behavior patterns in video sequences; training the behavior analysis model using a first data set; the trained behavior analysis model can identify various behavior events such as wandering, running, gathering, intrusion, etc.

The target recognition model and the behavior analysis model are integrated and optimized. Through joint training and multi-task learning, some underlying features of the target recognition model and the behavior analysis model are shared and their respective results are output respectively; thereby realizing complex functions on a single intelligent monitoring device, reducing computing resource consumption while improving model performance.

The performance indicators of the trained monitoring model are evaluated through the second data set, and the performance indicators include accuracy, recall rate and F1 value, and the model is adjusted and optimized according to the evaluation results to ensure that the model achieves the ideal effect in the real environment.

The adjusted and optimized monitoring model is converted into a form suitable for operation of embedded devices, and the converted monitoring model is deployed in the front-end intelligent monitoring device.

The working principle of the above technical solution is as follows: collect surveillance video clips containing diverse target types and behavior patterns from the Internet, annotate them in detail, and establish a target recognition and behavior analysis dataset that comprehensively covers different environmental factors (lighting, weather, crowd density, etc.). Divide the entire dataset into a training set (80%) and a validation set (20%); train the target recognition model using deep neural network architectures (such as YOLO, Faster R-CNN), which are good at real-time positioning and classification of different target objects from video frames. During the training process, the model will learn how to recognize and distinguish different target categories in complex backgrounds, such as pedestrians, vehicles, landmarks, etc. Build a behavior analysis model based on spatiotemporal convolutional network (STCNN), recurrent neural network (RNN) or long short-term memory network (LSTM), which can capture and understand the motion trajectory and behavior patterns in video sequences. Use the first dataset to train the behavior analysis model so that it can effectively identify various behavioral events, such as wandering, running, gathering, intrusion, etc. The target recognition model and the behavior analysis model are integrated, and through joint training and multi-task learning mechanisms, they are partially shared at the underlying feature level, so that the model can perform both target recognition and behavior analysis at the same time. The second data set is used to evaluate the performance of the trained monitoring model. The main reference indicators are accuracy, recall rate and F1 value. The model parameters are adjusted and optimized according to the evaluation results. After the model adjustment and optimization is completed, it is converted into a form suitable for the operation of embedded intelligent monitoring devices. Then, this trained, optimized and converted monitoring model is deployed on the front-end intelligent monitoring device, so that it can perform target recognition and behavior analysis tasks in real time in a hardware environment with low power consumption and limited computing power, thereby realizing intelligent monitoring management.

The effect of the above technical solution is: by obtaining surveillance video clips containing multiple target types and behavior patterns from the Internet, and covering various actual situations such as different lighting conditions, weather conditions, crowd density, target size and motion changes, the trained surveillance model has stronger generalization ability and can adapt to various complex surveillance scenarios. The target recognition model is trained using a deep neural network architecture (such as YOLO, Faster R-CNN), so that the monitoring equipment can accurately and quickly locate and identify multiple target objects from video frames, such as people, vehicles, specific markers, etc. A behavior analysis model suitable for behavior analysis (such as STCNN, RNN or LSTM) is constructed to capture and understand the behavior patterns in video sequences, and effectively identify and analyze various behavior events, such as wandering, running, gathering, intrusion, etc. The target recognition model and the behavior analysis model are integrated and optimized through joint training and multi-task learning, which realizes the sharing of underlying features, reduces the consumption of computing resources, and improves the performance of the model, so that a single intelligent monitoring device can simultaneously realize the composite functions of target recognition and behavior analysis. The second data set is used to evaluate the performance indicators of the trained monitoring model, such as accuracy, recall rate, and F1 value. According to the evaluation results, the model is adjusted and optimized in a targeted manner to ensure that the recognition and analysis effects of the model in the real environment meet the ideal standards. The adjusted and optimized monitoring model is converted into a form suitable for operation of embedded devices and can be directly deployed in front-end intelligent monitoring equipment, which is conducive to achieving low cost, low energy consumption, and high performance of the monitoring system, and further promoting the intelligent upgrade and popularization of monitoring equipment.

In one embodiment of the present invention, the monitoring device inputs the collected first monitoring data into the monitoring model, extracts targets and makes behavior judgments on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and a key frame to an edge computing node, including:

The front-end intelligent monitoring device collects the video stream through the built-in camera, and uses the collected video stream as the first monitoring data, and pre-processes the first monitoring data, wherein the pre-processing includes brightness/contrast adjustment, noise filtering, frame rate adaptation, and segmenting the continuous video stream into single-frame images;

Each frame of image is input into the deployed target recognition model, which performs target detection in real time based on the trained parameters and outputs target detection information, including the bounding box coordinates and category probability of each target. The main task of this stage is to accurately identify and extract the target object of focus from the complex background.

A continuous multi-frame image sequence is input into the behavior analysis model. By analyzing the motion trajectory and morphological changes of the target object in the time dimension, it is determined whether the target has abnormal behavior or a specific behavior pattern. For example, based on the position movement and posture changes of the same target over a period of time, behavioral events such as stagnation, rapid movement, and crossing areas can be identified.

Based on the results of target detection and behavior analysis, the first analysis result is generated by summarizing, and the first analysis result includes but is not limited to information such as target list, target status, and behavior event description. Frames where behavior events occur or have key information are selected and saved as key frames; after compression encoding, these key frames can retain important information and reduce the amount of data transmission.

The first analysis result and the selected key frame are encrypted and compressed, and the encrypted and compressed first analysis result and the selected key frame are transmitted to the edge computing node through a multi-channel transmission protocol.

The working principle of the above technical solution is as follows: the front-end intelligent monitoring device uses the built-in camera to collect video streams in real time as the first monitoring data, performs a series of preprocessing operations on the collected video, such as brightness/contrast adjustment to improve image quality, noise filtering to remove unnecessary interference, frame rate adaptation to ensure processing efficiency, and finally splits the continuous video stream into single-frame images for subsequent processing. The preprocessed single-frame images are sent to the deployed target recognition model for real-time processing. The target recognition model identifies and locates various types of targets in the image based on the trained parameters, outputs the specific bounding box coordinates and corresponding category probabilities of each target, and accurately extracts the target object of interest from the complex background. Continuous multi-frame images are input into the behavior analysis model, and the model analyzes the position, direction, speed and morphological changes of the target object at different time points to determine whether the target exhibits abnormal behavior or conforms to a specific behavior pattern. For example, by tracking the movement trajectory of the target over a period of time, behavioral events such as stagnation, rapid movement, and crossing the boundary can be identified. Combining the results of target detection and behavior analysis, the system generates a first analysis result, which contains detailed analysis information, such as a list of targets and their status, and a description of the behavioral events that occurred. At the same time, the system will select those frames containing important behavioral events or key information as key frames for storage and compress them to reduce the amount of data transmission but maintain the validity of key information. The first analysis results and the selected key frames will be encrypted and further compressed, and then safely and reliably transmitted to the edge computing node through a multi-channel transmission protocol. After receiving this data, the edge computing node can further make real-time decisions, store or integrate and analyze the data with other nodes, thereby achieving more efficient intelligent monitoring and early warning functions.

The effects of the above technical solutions are as follows: the monitoring equipment adopts the front-end intelligent processing method, which can pre-process and detect the target of the video stream in real time, effectively reduce the data transmission delay, and improve the response speed and real-time performance of the overall monitoring system; the pre-processing link ensures the image quality input to the target recognition model, which is conducive to the model to detect and identify the target object more accurately, whether in complex environments or low-light conditions, it can improve the accuracy of target extraction; the behavior analysis model can conduct in-depth analysis of the movement trajectory and morphological changes of the target object, effectively identify potential abnormal behaviors or specific behavior patterns, and enhance the intelligent prediction and early warning capabilities of the monitoring system, which is crucial for security prevention and event response. By selecting key frames for transmission and storage, the amount of invalid data transmission is greatly reduced, network bandwidth and storage space are saved, and the pressure on back-end computing is also reduced, achieving efficient resource utilization and energy conservation and emission reduction. The first analysis result and key frame are encrypted and compressed and transmitted to the edge computing node, which helps to realize the integration of cloud computing and edge computing, so that part of the analysis and decision-making process can be completed close to the data source, speeding up the decision-making speed and improving the overall performance of the system. Data encryption transmission ensures the security of sensitive information, effectively prevents data from being illegally intercepted and tampered during transmission, and improves the data security level of the monitoring system.

In one embodiment of the present invention, the edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules to obtain second monitoring data, and uploads the second monitoring data to the cloud server; the second monitoring data includes abnormal events or important information, including:

The edge computing node receives the first analysis result and key frame data sent from the front-end intelligent monitoring device, and performs decryption and decompression;

According to preset rules, the target information and behavior events in the first analysis result are preliminarily screened to eliminate false positives and data that do not meet the conditions, and verify whether the content of the key frame matches the analysis result;

The filtered data is further analyzed and optimized through parallel processing algorithms, such as further refining behavior recognition and enhancing target tracking effects. And according to the preset business rules and security policies, abnormal events are identified and marked from the optimized data, such as intrusion behavior, fire warning, abnormal crowd gathering, etc. And the information of key time and space points is extracted to obtain the second monitoring data.

If there are multiple monitoring sources, the edge computing node is used to fuse and correlate cross-source data to integrate related events or information from different sources;

The second monitoring data is compressed and encrypted, and the packaged second monitoring data is prioritized according to the severity of the abnormal event or the urgency level of important information to ensure that high-priority data can be uploaded to the cloud server as soon as possible; and the compressed and encrypted second monitoring data is transmitted to the cloud server according to priority through a multi-channel transmission protocol.

The working principle of the above technical solution is as follows: the edge computing node receives the first analysis result and the encrypted and compressed key frame data sent by the front-end intelligent monitoring device, first performs decryption and decompression operations to restore the original analysis result and key image information; the first analysis result is screened using preset rules to remove possible false alarms or targets and behavior events that do not meet the requirements of the application scenario. At the same time, the consistency between the key frame content and the analysis result is verified to ensure the accuracy of the data. The screened data is analyzed and optimized at a deeper level using parallel processing algorithms, such as improving the continuity and stability of the target trajectory through improved target tracking methods, and more finely identifying and distinguishing behavior patterns to achieve higher recognition accuracy and lower missed detection rates. According to predefined business rules and security policies, the optimized data is deeply mined to identify and mark real abnormal events, such as security intrusions, fire alarms, etc., or abnormal crowd dynamics in public places, etc., and the time and space key point information of these events is extracted to form the second monitoring data. When there are multiple monitoring sources, the edge computing node is responsible for integrating relevant data from different sources, and by performing correlation analysis on these data, potential related events or clues are revealed, thereby providing a more comprehensive situational awareness. According to the severity of the abnormal event or the urgency level of important information, the second monitoring data is prioritized, compressed and encrypted, and then uploaded to the cloud server in order of priority. This mechanism ensures that data in emergency situations can be quickly transmitted so that the cloud can provide higher-level decision support or linkage response.

The effect of the above technical solution is: the edge computing node performs preliminary screening and verification by decrypting and decompressing the received first analysis results and key frame data, effectively avoiding the transmission of invalid data and false alarms, and improving data processing efficiency and accuracy. The parallel processing algorithm re-analyzes and optimizes the screened data, such as refining behavior recognition and enhancing target tracking effects, making the understanding and analysis of the monitoring scene more in-depth and accurate, and can more accurately capture and record the dynamic behavior of the target. According to the preset business rules and security policies, the system can identify and mark abnormal events such as intrusion, fire warning, abnormal crowd gathering, etc. from the optimized data in real time, and can quickly extract key time and space information to provide immediate and effective data support for subsequent decision-making. If there are multiple monitoring sources, the edge computing node can perform cross-source data fusion and correlation analysis, integrating different sources but related events or information together to form a more comprehensive and three-dimensional monitoring view, which helps to discover hidden correlations or trends. The second monitoring data is prioritized according to the severity of the abnormal event or the emergency level of important information, and is efficiently transmitted after encryption and compression through a multi-channel transmission protocol to ensure that emergency information can be quickly delivered to the cloud server, which is conducive to faster decision-making and taking corresponding actions. The data is screened and optimized through edge computing nodes, and most of the computing tasks are completed on the edge, which reduces the computing pressure of the cloud server, realizes resource optimization configuration and load balancing, and improves the stability and response speed of the entire monitoring system. The second monitoring data is compressed and encrypted to ensure the security and privacy protection of the data during transmission, and reduce the risk of data leakage.

In one embodiment of the present invention, the cloud server stores the received second monitoring data, further analyzes and mines the second monitoring data by performing a big data analysis algorithm, generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel, including:

The cloud server receives the second monitoring data uploaded by the edge computing node, decrypts and decompresses the second monitoring data, and stores the second monitoring data in different storage spaces according to the priority order;

The cloud server re-preprocesses the second monitoring data stored in different storage spaces, and the re-preprocessing includes removing duplicates, filling missing values, detecting and processing outliers; and performs feature engineering processing on the re-preprocessed second monitoring data; the feature engineering processing includes extracting key features, constructing spatiotemporal indexes, and performing data standardization or normalization.

The pre-processed data is deeply analyzed through a machine learning algorithm to obtain a second analysis result; the second analysis result includes potential patterns, trends and abnormal situations; and based on spatiotemporal series analysis, cluster analysis and association rule analysis algorithms, a third analysis result is obtained; the third analysis result is the deep-level rules and key information hidden in a large amount of monitoring data.

If it is a specific scenario (such as security, transportation, production, etc.), targeted analysis is performed through domain knowledge and expert systems to obtain a fourth analysis result, which is the location of the risk area and important events.

Based on the various analysis results, a comprehensive situation model is constructed, and the regional situation of the comprehensive situation model is evaluated to obtain a fourth analysis result, wherein the region includes the global and local regions, and the situation includes the security situation and the operation status;

The first to fourth analysis results are summarized and a comprehensive situation report is generated, which includes an overview of the current situation, problem findings, development trend forecasts, and improvement suggestions; the comprehensive situation report is pushed to relevant managers through various methods, including emails, text messages, and internal communication software; and the relevant managers take corresponding actions after receiving the comprehensive situation report.

The working principle of the above technical solution is as follows: the cloud server first receives the second monitoring data that has been screened, optimized and encrypted by the edge computing node, and decrypts and decompresses it. The data is stored in different storage spaces according to the priority order when it is uploaded, so that it can be accessed and processed on demand later. The second monitoring data stored in each space is preprocessed again, including deduplication, missing value filling, and outlier detection and processing to ensure data quality. Feature engineering processing is performed to convert the original data into a feature set suitable for machine learning algorithms, such as extracting key features, establishing spatiotemporal indexes, and performing data standardization or normalization. The machine learning algorithm is used to conduct in-depth analysis of the preprocessed data to explore potential patterns, trends and abnormal situations, and form the second analysis results. Using means such as spatiotemporal sequence analysis, cluster analysis and association rule analysis, deep-level laws and key information are mined from a large amount of monitoring data to form the third analysis results. For specific application scenarios (such as security, transportation, production, etc.), targeted analysis is carried out in combination with domain knowledge and expert systems to obtain the fourth analysis results, which may contain specific information such as risk area positioning and important event identification. Based on the analysis results at all levels, a comprehensive situation model is constructed to comprehensively evaluate the security situation and operation status of the global and local areas, forming a more detailed fourth analysis result. The analysis results of the four stages are summarized to generate a complete comprehensive situation report, which covers multiple aspects such as current situation overview, problem discovery, development trend prediction and improvement suggestions. The comprehensive situation report is pushed to relevant managers in a timely manner through various channels such as email, text messages, and internal communication software, so that they can make corresponding decisions and management actions based on the content of the report. For example: Assume that there is a monitoring system for a large shopping mall: The comprehensive situation report first summarizes the overall security status of the shopping mall in the past month, including the total number of people entering and leaving, the average stay time, the peak period, and the frequency of violations (such as tailing, detention, and lost items) in the monitoring area. Through spatiotemporal sequence analysis and behavior recognition algorithms, the report found that at the entrance of the parking lot in the northeast corner of the shopping mall, there has been an increase in vehicles entering unauthorized areas in the evening recently, and they are often accompanied by unidentified people following customers into the parking lot. Based on historical data and machine learning model predictions, if no measures are taken, it is expected that such violations may continue to increase during the evening peak period in the next two weeks, which may pose a potential threat to customer safety. It is recommended to strengthen security forces in problem areas, especially to add security patrols during the evening peak hours; at the same time, upgrade the intelligent monitoring system in the area, such as adding high-definition cameras, installing license plate recognition systems, and connecting to the public security system to strengthen the tracking and control of suspicious vehicles and personnel. After receiving the comprehensive situation report, the security management team can take the following actions based on the report content: immediately adjust and deploy security forces to ensure that there are enough personnel to patrol the problem area; communicate with the technical team to implement the upgrade and networking of intelligent monitoring equipment as soon as possible; cooperate with the local police to share monitoring information and jointly combat possible illegal activities; convey relevant information to shopping center employees, reminding them to pay attention to personal safety and customer service during peak hours, especially to strengthen vigilance in problem areas.

The effect of the above technical solution is: by receiving and hierarchically storing the second monitoring data uploaded by the edge computing node through the cloud server, large-scale, real-time data collection and efficient storage can be achieved, data loss can be avoided, and it is convenient for rapid retrieval and utilization according to priority. The received monitoring data is decrypted, decompressed and preprocessed, which effectively solves the problems of data redundancy, missing and abnormality, and improves the accuracy and reliability of the analysis results. Feature engineering processing helps to extract the key features of the monitoring data, build a spatiotemporal index, structure the multi-dimensional and time series data, and ensure the comparability of data from different sources and scales through standardization or normalization. Using machine learning algorithms to conduct in-depth analysis of the preprocessed data can reveal potential patterns, trends and abnormal situations, warn of potential risks in advance, and help improve the foresight and initiative of management decisions. In specific industry scenarios, combined with domain knowledge and expert systems for analysis, the location and important events of high-risk areas can be accurately identified, providing managers with specific response strategy basis. A variety of analysis methods are comprehensively applied to build a comprehensive situation model, objectively and comprehensively evaluate the security situation and operating status of the overall and local areas, and form an intuitive and easy-to-understand situation report. The automatically generated comprehensive situation report includes an overview of the current situation, problem findings, development trend forecasts, and improvement suggestions, and is quickly pushed to relevant managers in various forms to accelerate the decision-making process and improve response speed.

One embodiment of the present invention, as shown in FIG2 , is an intelligent monitoring and management system, the system comprising:

Model training module: trains the monitoring model in the monitoring scenario through deep learning algorithms, the monitoring model includes a target recognition model and a behavior analysis model, and deploys the trained monitoring model on the front-end intelligent monitoring device;

Data transmission module: The monitoring device inputs the collected first monitoring data into the monitoring model, performs target extraction and behavior judgment on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and key frames to the edge computing node;

Data upload module: the edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules, obtains second monitoring data, and uploads the second monitoring data to the cloud server; the second monitoring data includes abnormal events or important information;

Result feedback module: The cloud server stores the received second monitoring data, and further analyzes and mines the second monitoring data by performing a big data analysis algorithm, and generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel.

The working principle of the above technical solution is: use the deep learning algorithm to train the data set in the monitoring scene and establish a monitoring model, which includes a target recognition model and a behavior analysis model. The target recognition model is used to accurately identify various targets (such as people, vehicles, etc.) from the video stream, and the behavior analysis model is responsible for analyzing the behavioral characteristics and activity patterns of the target; the trained monitoring model is deployed on the front-end intelligent monitoring device. The monitoring device captures the monitoring screen in real time as the first monitoring data and inputs it into the monitoring model for real-time analysis. The monitoring model performs target extraction in real time, identifies and locates each target in the monitoring screen, and determines whether the target's behavior has abnormalities or important features based on the behavior analysis model to form a preliminary first analysis result. At the same time, the key frame (i.e., the frame containing important information) will be extracted and sent to the edge computing node together with the first analysis result. The edge computing node performs secondary screening and optimization of the first analysis results and key frames transmitted from the front end according to the preset rules, eliminates redundant information, enhances the quality of effective information, and then generates more refined and targeted second monitoring data. The second monitoring data mainly focuses on abnormal events or important information in the monitoring scene, reduces the amount of original data transmission, and improves data processing efficiency. After receiving the second monitoring data uploaded by the edge computing node, the cloud server will store and manage it for a long time and on a large scale; it will use big data analysis algorithms to deeply mine and analyze these data to find potential correlations, patterns and trends, thereby forming a comprehensive situation report covering the overall monitoring situation. Finally, the cloud server will feed back the generated comprehensive situation report to relevant personnel through appropriate channels for decision-making reference, emergency response or daily management.

The effects of the above technical solution are: the target recognition model and behavior analysis model trained by the deep learning algorithm can achieve high-precision target detection and behavior understanding, improve the intelligence level of the monitoring system, and accurately capture the detailed information in the monitoring scene; the front-end intelligent monitoring equipment is equipped with a trained model, which can realize real-time target extraction and behavior judgment, quickly discover potential safety hazards or abnormal situations, and help improve the response speed to emergencies. By introducing edge computing nodes, a large amount of monitoring data generated by the front-end equipment can be processed locally and quickly, and truly valuable data (second monitoring data) can be screened out, and invalid data transmission can be reduced. At the same time, the network bandwidth pressure is reduced, and the timeliness and accuracy of data processing are improved. The second monitoring data pre-processed by the edge computing node is uploaded to the cloud server, so that the cloud platform does not need to process all the original data, but focuses on in-depth analysis of key abnormal events or important information, which improves the effective use of cloud computing resources. The cloud server uses big data analysis technology to deeply mine the second monitoring data, which can reveal the patterns and trends hidden in the massive monitoring information, form a comprehensive situation report, and help managers fully grasp the overall security situation and activity rules of the monitoring area. Comprehensive situation reports can provide strong decision-making support for relevant departments and personnel, promote rapid and scientific decision-making, and take timely and effective actions to deal with various security issues or emergencies.

In one embodiment of the present invention, the model training module includes:

Dataset division module: Obtain public surveillance video clips containing multiple target types and behavior patterns from the Internet through the API interface, annotate the surveillance clips, obtain target recognition and behavior analysis data sets, and use 80% of the data sets as the first data set and 20% as the second data set; the data sets cover but are not limited to different lighting conditions, weather conditions, crowd density, and target size and action changes;

Data input module: input the labeled first data set into a deep neural network architecture (such as YOLO, FasterR-CNN), and train the target recognition model through the deep neural network architecture; the trained target recognition model can accurately locate and classify different target objects from the original video frames, such as people, vehicles, specific markers, etc.

Model building module: constructing a behavior analysis model suitable for behavior analysis (such as a spatiotemporal convolutional network (STCNN), a recurrent neural network (RNN), or a long short-term memory network (LSTM)), wherein the behavior analysis model is used to capture and understand motion trajectories and behavior patterns in video sequences; training the behavior analysis model using a first data set; the trained behavior analysis model can identify various behavioral events such as wandering, running, gathering, and intrusion.

Fusion optimization module: Fusion and optimization of the target recognition model and the behavior analysis model. Through joint training and multi-task learning, some underlying features of the target recognition model and the behavior analysis model are shared and their respective results are output respectively; thereby realizing complex functions on a single intelligent monitoring device, reducing computing resource consumption while improving model performance.

Performance evaluation module: The performance indicators of the trained monitoring model are evaluated through the second data set, and the performance indicators include accuracy, recall rate and F1 value, and the model is adjusted and optimized according to the evaluation results to ensure that the model achieves the ideal effect in the real environment.

Model conversion module: converts the adjusted and optimized monitoring model into a form suitable for embedded devices, and deploys the converted monitoring model in the front-end intelligent monitoring device.

The working principle of the above technical solution is as follows: collect surveillance video clips containing diverse target types and behavior patterns from the Internet, annotate them in detail, and establish a target recognition and behavior analysis dataset that comprehensively covers different environmental factors (lighting, weather, crowd density, etc.). Divide the entire dataset into a training set (80%) and a validation set (20%); train the target recognition model using deep neural network architectures (such as YOLO, Faster R-CNN), which are good at real-time positioning and classification of different target objects from video frames. During the training process, the model will learn how to recognize and distinguish different target categories in complex backgrounds, such as pedestrians, vehicles, landmarks, etc. Build a behavior analysis model based on spatiotemporal convolutional network (STCNN), recurrent neural network (RNN) or long short-term memory network (LSTM), which can capture and understand the motion trajectory and behavior patterns in video sequences. Use the first dataset to train the behavior analysis model so that it can effectively identify various types of behavioral events, such as wandering, running, gathering, intrusion, etc. The target recognition model and the behavior analysis model are integrated, and through joint training and multi-task learning mechanisms, they are partially shared at the underlying feature level, so that the model can perform both target recognition and behavior analysis at the same time. The performance of the trained monitoring model is evaluated using the second data set. The main reference indicators are accuracy, recall rate and F1 value. The model parameters are adjusted and optimized according to the evaluation results. After the model adjustment and optimization is completed, it is converted into a form suitable for the operation of embedded intelligent monitoring devices. Then, this trained, optimized and converted monitoring model is deployed on the front-end intelligent monitoring device, so that it can perform target recognition and behavior analysis tasks in real time in a hardware environment with low power consumption and limited computing power, thereby realizing intelligent monitoring management.

The effect of the above technical solution is: by obtaining surveillance video clips containing multiple target types and behavior patterns from the Internet, and covering various actual situations such as different lighting conditions, weather conditions, crowd density, target size and motion changes, the trained monitoring model has stronger generalization ability and can adapt to various complex monitoring scenarios. The target recognition model is trained using a deep neural network architecture (such as YOLO, Faster R-CNN), so that the monitoring device can accurately and quickly locate and identify multiple target objects from video frames, such as people, vehicles, specific markers, etc. A behavior analysis model suitable for behavior analysis (such as STCNN, RNN or LSTM) is constructed to capture and understand the behavior patterns in video sequences, and effectively identify and analyze various behavior events, such as wandering, running, gathering, intrusion, etc. The target recognition model and the behavior analysis model are integrated and optimized through joint training and multi-task learning, which realizes the sharing of underlying features, reduces the consumption of computing resources, and improves the performance of the model, so that a single intelligent monitoring device can simultaneously realize the composite functions of target recognition and behavior analysis. The second data set is used to evaluate the performance indicators of the trained monitoring model, such as accuracy, recall rate, and F1 value. According to the evaluation results, the model is adjusted and optimized in a targeted manner to ensure that the recognition and analysis effects of the model in the real environment meet the ideal standards. The adjusted and optimized monitoring model is converted into a form suitable for operation of embedded devices and can be directly deployed in front-end intelligent monitoring equipment, which is conducive to achieving low cost, low energy consumption and high performance of the monitoring system, and further promoting the intelligent upgrade and popularization of monitoring equipment.

In one embodiment of the present invention, the data transmission module includes:

Data acquisition module: The front-end intelligent monitoring device acquires the video stream through the built-in camera, and uses the acquired video stream as the first monitoring data, and pre-processes the first monitoring data. The pre-processing includes brightness/contrast adjustment, noise filtering, frame rate adaptation, and segmenting the continuous video stream into single-frame images;

Object detection module: Each frame of image is input into the deployed object recognition model, which performs object detection in real time based on the trained parameters and outputs object detection information; the object detection information includes the bounding box coordinates and category probability of each object. The main task of this stage is to accurately identify and extract the target object of focus from the complex background.

Behavior judgment module: inputs a continuous multi-frame image sequence into the behavior analysis model, and judges whether the target has abnormal behavior or specific behavior patterns by analyzing the motion trajectory and morphological changes of the target object in the time dimension; for example, for the position movement and posture changes of the same target over a period of time, behavioral events such as stagnation, rapid movement, and crossing areas can be identified.

Saving module: Based on the results of target detection and behavior analysis, the first analysis result is generated by summarizing, including but not limited to target list, target status, behavior event description and other information. Frames where behavior events occur or have key information are selected and saved as key frames; after compression encoding, these key frames can retain important information and reduce the amount of data transmission.

Encryption and compression module: encrypt and compress the first analysis result and the selected key frame, and transmit the encrypted and compressed first analysis result and the selected key frame to the edge computing node through a multi-channel transmission protocol.

The working principle of the above technical solution is as follows: the front-end intelligent monitoring device uses the built-in camera to collect video streams in real time as the first monitoring data, performs a series of preprocessing operations on the collected video, such as brightness/contrast adjustment to improve image quality, noise filtering to remove unnecessary interference, frame rate adaptation to ensure processing efficiency, and finally splits the continuous video stream into single-frame images for subsequent processing. The preprocessed single-frame images are sent to the deployed target recognition model for real-time processing. The target recognition model identifies and locates various types of targets in the image based on the trained parameters, outputs the specific bounding box coordinates and corresponding category probabilities of each target, and accurately extracts the target object of interest from the complex background. Continuous multi-frame images are input into the behavior analysis model, and the model analyzes the position, direction, speed and morphological changes of the target object at different time points to determine whether the target exhibits abnormal behavior or conforms to a specific behavior pattern. For example, by tracking the movement trajectory of the target over a period of time, behavioral events such as stagnation, rapid movement, and crossing the boundary can be identified. Combining the results of target detection and behavior analysis, the system generates a first analysis result, which contains detailed analysis information, such as a list of targets and their status, and a description of the behavioral events that occurred. At the same time, the system will select those frames containing important behavioral events or key information as key frames for storage and compress them to reduce the amount of data transmission but maintain the validity of key information. The first analysis results and the selected key frames will be encrypted and further compressed, and then safely and reliably transmitted to the edge computing node through a multi-channel transmission protocol. After receiving this data, the edge computing node can further make real-time decisions, store or integrate and analyze the data with other nodes, thereby achieving more efficient intelligent monitoring and early warning functions.

The effects of the above technical solutions are as follows: the monitoring equipment adopts the front-end intelligent processing method, which can pre-process and detect the target of the video stream in real time, effectively reduce the data transmission delay, and improve the response speed and real-time performance of the overall monitoring system; the pre-processing link ensures the image quality input to the target recognition model, which is conducive to the model to detect and identify the target object more accurately, whether in complex environments or low-light conditions, it can improve the accuracy of target extraction; the behavior analysis model can conduct in-depth analysis of the movement trajectory and morphological changes of the target object, effectively identify potential abnormal behaviors or specific behavior patterns, and enhance the intelligent prediction and early warning capabilities of the monitoring system, which is crucial for security prevention and event response. By selecting key frames for transmission and storage, the amount of invalid data transmission is greatly reduced, network bandwidth and storage space are saved, and the pressure on back-end computing is also reduced, achieving efficient resource utilization and energy conservation and emission reduction. The first analysis result and key frame are encrypted and compressed and transmitted to the edge computing node, which helps to realize the integration of cloud computing and edge computing, so that part of the analysis and decision-making process can be completed close to the data source, speeding up the decision-making speed and improving the overall performance of the system. Data encryption transmission ensures the security of sensitive information, effectively prevents data from being illegally intercepted and tampered with during transmission, and improves the data security level of the monitoring system.

In one embodiment of the present invention, the data uploading module includes:

Decryption and decompression module: The edge computing node receives the first analysis result and key frame data sent by the front-end intelligent monitoring device, and performs decryption and decompression;

Preliminary screening module: Preliminary screening of target information and behavior events in the first analysis result according to preset rules, eliminating false positives and data that do not meet the conditions, and verifying whether the content of the key frame matches the analysis result;

Secondary optimization module: The filtered data is analyzed and optimized again through parallel processing algorithms, such as further refining behavior recognition, enhancing target tracking effects, etc. And according to the preset business rules and security policies, abnormal events are identified and marked from the optimized data, such as intrusion behavior, fire warning, abnormal crowd gathering, etc. And the information of key time and space points is extracted to obtain the second monitoring data.

Integration module: If there are multiple monitoring sources, the edge computing node is used to fuse and correlate cross-source data, and integrate related events or information from different sources;

Priority sorting module: compress and encrypt the second monitoring data, prioritize the packaged second monitoring data according to the severity of the abnormal event or the urgency level of important information, and ensure that high-priority data can be uploaded to the cloud server as soon as possible; and transmit the compressed and encrypted second monitoring data to the cloud server according to priority through a multi-channel transmission protocol.

The working principle of the above technical solution is as follows: the edge computing node receives the first analysis result and the encrypted and compressed key frame data sent by the front-end intelligent monitoring device, first performs decryption and decompression operations to restore the original analysis result and key image information; the first analysis result is screened using preset rules to remove possible false alarms or targets and behavior events that do not meet the requirements of the application scenario. At the same time, the consistency between the key frame content and the analysis result is verified to ensure the accuracy of the data. The screened data is analyzed and optimized at a deeper level using parallel processing algorithms, such as improving the continuity and stability of the target trajectory through improved target tracking methods, and more finely identifying and distinguishing behavior patterns to achieve higher recognition accuracy and lower missed detection rates. According to predefined business rules and security policies, the optimized data is deeply mined to identify and mark real abnormal events, such as security intrusions, fire alarms, etc., or abnormal crowd dynamics in public places, etc., and the time and space key point information of these events is extracted to form the second monitoring data. When there are multiple monitoring sources, the edge computing node is responsible for integrating relevant data from different sources, and by performing correlation analysis on these data, potential related events or clues are revealed, thereby providing a more comprehensive situational awareness. According to the severity of the abnormal event or the urgency level of important information, the second monitoring data is prioritized, compressed and encrypted, and then uploaded to the cloud server in order of priority. This mechanism ensures that data in emergency situations can be quickly transmitted so that the cloud can provide higher-level decision support or linkage response.

The effect of the above technical solution is: the edge computing node performs preliminary screening and verification by decrypting and decompressing the received first analysis results and key frame data, effectively avoiding the transmission of invalid data and false alarms, and improving data processing efficiency and accuracy. The parallel processing algorithm re-analyzes and optimizes the screened data, such as refining behavior recognition and enhancing target tracking effects, making the understanding and analysis of the monitoring scene more in-depth and accurate, and can more accurately capture and record the dynamic behavior of the target. According to the preset business rules and security policies, the system can identify and mark abnormal events such as intrusion, fire warning, abnormal crowd gathering, etc. from the optimized data in real time, and can quickly extract key time and space information to provide immediate and effective data support for subsequent decision-making. If there are multiple monitoring sources, the edge computing node can perform cross-source data fusion and correlation analysis, integrating events or information from different sources but related, forming a more comprehensive and three-dimensional monitoring view, which helps to discover hidden correlations or trends. The second monitoring data is prioritized according to the severity of the abnormal event or the emergency level of important information, and is efficiently transmitted after encryption and compression through a multi-channel transmission protocol to ensure that emergency information can be quickly delivered to the cloud server, which is conducive to faster decision-making and taking corresponding actions. The data is screened and optimized through edge computing nodes, and most of the computing tasks are completed on the edge, which reduces the computing pressure of the cloud server, realizes resource optimization configuration and load balancing, and improves the stability and response speed of the entire monitoring system. The second monitoring data is compressed and encrypted to ensure the security and privacy protection of the data during transmission, and reduce the risk of data leakage.

In one embodiment of the present invention, the result feedback module includes:

Space allocation module: the cloud server receives the second monitoring data uploaded by the edge computing node, decrypts and decompresses the second monitoring data, and stores the second monitoring data in different storage spaces according to the priority order;

Preprocessing module: The cloud server re-preprocesses the second monitoring data stored in different storage spaces, and the re-preprocessing includes removing duplicates, filling missing values, detecting and processing outliers; and performs feature engineering on the re-preprocessed second monitoring data; the feature engineering includes extracting key features, constructing spatiotemporal indexes, and performing data standardization or normalization.

Analysis result module: The pre-processed data is deeply analyzed through machine learning algorithms to obtain a second analysis result; the second analysis result includes potential patterns, trends and abnormal situations; and based on spatiotemporal sequence analysis, cluster analysis and association rule analysis algorithms, a third analysis result is obtained; the third analysis result is the deep-level rules and key information hidden in a large amount of monitoring data.

The third analysis module: If it is a specific scenario (such as security, transportation, production, etc.), targeted analysis is performed through domain knowledge and expert systems to obtain the fourth analysis result, which is the location of the risk area and important events.

Fourth analysis module: based on various analysis results, construct a comprehensive situation model, evaluate the regional situation of the comprehensive situation model, and obtain a fourth analysis result, wherein the region includes the global and local areas, and the situation includes the security situation and the operation status;

Report feedback module: summarize the first to fourth analysis results and generate a comprehensive situation report, which includes an overview of the current situation, problem findings, development trend forecasts and improvement suggestions; push the comprehensive situation report to relevant managers through various methods, including emails, text messages, and internal communication software; relevant managers take corresponding actions after receiving the comprehensive situation report.

The working principle of the above technical solution is as follows: the cloud server first receives the second monitoring data that has been screened, optimized and encrypted by the edge computing node, and decrypts and decompresses it. The data is stored in different storage spaces according to the priority order when it is uploaded, so that it can be accessed and processed on demand later. The second monitoring data stored in each space is preprocessed again, including deduplication, missing value filling, and outlier detection and processing to ensure data quality. Feature engineering processing is performed to convert the original data into a feature set suitable for machine learning algorithms, such as extracting key features, establishing spatiotemporal indexes, and performing data standardization or normalization. The machine learning algorithm is used to conduct in-depth analysis of the preprocessed data to explore potential patterns, trends and abnormal situations, and form the second analysis results. Using means such as spatiotemporal sequence analysis, cluster analysis and association rule analysis, deep-level laws and key information are mined from a large amount of monitoring data to form the third analysis results. For specific application scenarios (such as security, transportation, production, etc.), targeted analysis is carried out in combination with domain knowledge and expert systems to obtain the fourth analysis results, which may contain specific information such as risk area positioning and important event identification. Based on the analysis results at all levels, a comprehensive situation model is constructed to conduct a comprehensive assessment of the security situation and operating status of the global and local areas, forming a more detailed fourth analysis result. The analysis results of the four stages are summarized to generate a complete comprehensive situation report, which covers multiple aspects such as current situation overview, problem discovery, development trend forecast, and improvement suggestions. The comprehensive situation report is pushed to relevant managers in a timely manner through various channels such as email, text messages, and internal communication software, so that they can make corresponding decisions and management actions based on the content of the report.

The effect of the above technical solution is: by receiving and hierarchically storing the second monitoring data uploaded by the edge computing node through the cloud server, large-scale, real-time data collection and efficient storage can be achieved, data loss can be avoided, and it is convenient for rapid retrieval and utilization according to priority. The received monitoring data is decrypted, decompressed and preprocessed, which effectively solves the problems of data redundancy, missing and abnormality, and improves the accuracy and reliability of the analysis results. Feature engineering processing helps to extract the key features of the monitoring data, build a spatiotemporal index, structure the multi-dimensional and time series data, and ensure the comparability of data from different sources and scales through standardization or normalization. Using machine learning algorithms to conduct in-depth analysis of the preprocessed data can reveal potential patterns, trends and abnormal situations, warn of potential risks in advance, and help improve the foresight and initiative of management decisions. In specific industry scenarios, combined with domain knowledge and expert systems for analysis, the location and important events of high-risk areas can be accurately identified, providing managers with specific response strategy basis. A variety of analysis methods are comprehensively applied to build a comprehensive situation model, objectively and comprehensively evaluate the security situation and operating status of the overall and local areas, and form an intuitive and easy-to-understand situation report. The automatically generated comprehensive situation report includes an overview of the current situation, problem findings, development trend forecasts, and improvement suggestions, and is quickly pushed to relevant managers in various forms to accelerate the decision-making process and improve response speed.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims (10)

1. An intelligent monitoring and management method, characterized in that the method comprises: The monitoring model in the monitoring scenario is trained by a deep learning algorithm, wherein the monitoring model includes a target recognition model and a behavior analysis model, and the trained monitoring model is deployed on a front-end intelligent monitoring device; The monitoring device inputs the collected first monitoring data into the monitoring model, performs target extraction and behavior judgment on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and the key frame to the edge computing node; The edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules to obtain second monitoring data, and uploads the second monitoring data to the cloud server; The cloud server stores the received second monitoring data, further analyzes and mines the second monitoring data by performing a big data analysis algorithm, generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel. 2. According to claim 1, an intelligent monitoring management method is characterized in that the monitoring model in the monitoring scenario is trained by a deep learning algorithm, the monitoring model includes a target recognition model and a behavior analysis model, and the trained monitoring model is deployed on a front-end intelligent monitoring device, including: Obtaining public surveillance video clips containing multiple target types and behavior patterns from the Internet through an API interface, annotating the surveillance clips, obtaining target recognition and behavior analysis data sets, and using 80% of the data sets as the first data set and 20% as the second data set; Inputting the labeled first data set into a deep neural network architecture, and training an object recognition model through the deep neural network architecture; Constructing a behavior analysis model suitable for behavior analysis, and training the behavior analysis model using a first data set; The target recognition model and the behavior analysis model are integrated and optimized, and some underlying features of the target recognition model and the behavior analysis model are shared through joint training and multi-task learning, and their respective results are outputted; Evaluate the performance indicators of the trained monitoring model using the second data set, and adjust and optimize the model according to the evaluation results; The adjusted and optimized monitoring model is converted and deployed in the front-end intelligent monitoring device. 3. According to claim 1, an intelligent monitoring management method is characterized in that the monitoring device inputs the collected first monitoring data into the monitoring model, extracts targets and makes behavior judgments on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and key frames to the edge computing node, including: The front-end intelligent monitoring device collects the video stream through the built-in camera, uses the collected video stream as the first monitoring data, pre-processes the first monitoring data, and divides the continuous video stream into single-frame images; Input each frame of the image into the deployed target recognition model, which performs target detection in real time based on the trained parameters and outputs target detection information; Input a continuous multi-frame image sequence into the behavior analysis model, and determine whether the target has abnormal behavior or specific behavior patterns by analyzing the target object's motion trajectory and morphological changes in the time dimension; According to the results of target detection and behavior analysis, a first analysis result is generated by summarizing, and frames where behavior events occur or frames with key information are selected and saved as key frames; The first analysis result and the selected key frame are encrypted and compressed, and the encrypted and compressed first analysis result and the selected key frame are transmitted to the edge computing node through a multi-channel transmission protocol. 4. According to claim 1, an intelligent monitoring management method is characterized in that the edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules to obtain second monitoring data, and uploads the second monitoring data to the cloud server; the second monitoring data includes abnormal events or important information, including: The edge computing node receives the first analysis result and key frame data sent from the front-end intelligent monitoring device, and performs decryption and decompression; According to preset rules, the target information and behavior events in the first analysis result are preliminarily screened to eliminate false positives and data that do not meet the conditions, and verify whether the content of the key frame matches the analysis result; The filtered data is analyzed and optimized again through parallel processing algorithms, and abnormal events are identified and marked from the optimized data according to preset business rules and security policies, and information of key time and space points is extracted to obtain the second monitoring data; If there are multiple monitoring sources, the edge computing node is used to fuse and correlate cross-source data to integrate related events or information from different sources; The second monitoring data is compressed and encrypted, the packaged second monitoring data is prioritized according to the severity of the abnormal event or the urgency level of the important information, and the compressed and encrypted second monitoring data is transmitted to the cloud server according to the priority through a multi-channel transmission protocol. 5. According to claim 1, the intelligent monitoring management method is characterized in that the cloud server stores the received second monitoring data, further analyzes and mines the second monitoring data by performing a big data analysis algorithm, generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel, including: The cloud server receives the second monitoring data uploaded by the edge computing node, decrypts and decompresses the second monitoring data, and stores the second monitoring data in different storage spaces according to the priority order; The cloud server re-preprocesses the second monitoring data stored in different storage spaces, and performs feature engineering processing on the re-preprocessed second monitoring data; The pre-processed data is analyzed in depth through a machine learning algorithm to obtain a second analysis result; and a third analysis result is obtained based on spatiotemporal sequence analysis, cluster analysis, and association rule analysis algorithms; If it is a specific scenario, targeted analysis is performed through domain knowledge and expert systems to obtain the fourth analysis result. Based on various analysis results, a comprehensive situation model is constructed, and the regional situation of the comprehensive situation model is evaluated to obtain the fourth analysis result; The first to fourth analysis results are summarized and a comprehensive situation report is generated. The comprehensive situation report is pushed to relevant managers in various ways. After receiving the comprehensive situation report, the relevant managers perform corresponding processing. 6. An intelligent monitoring and management system, characterized in that the system comprises: Model training module: trains the monitoring model in the monitoring scenario through deep learning algorithms, the monitoring model includes a target recognition model and a behavior analysis model, and deploys the trained monitoring model on the front-end intelligent monitoring device; Data transmission module: The monitoring device inputs the collected first monitoring data into the monitoring model, performs target extraction and behavior judgment on the collected first monitoring data in real time through the monitoring model, obtains a first analysis result, and sends the first analysis result and key frames to the edge computing node; Data upload module: the edge computing node performs secondary screening and optimization on the received first analysis results and key frames according to preset rules, obtains second monitoring data, and uploads the second monitoring data to the cloud server; Result feedback module: The cloud server stores the received second monitoring data, and further analyzes and mines the second monitoring data by performing a big data analysis algorithm, and generates a comprehensive situation report, and feeds back the comprehensive situation report to relevant personnel. 7. The intelligent monitoring and management system according to claim 6, characterized in that the model training module comprises: Dataset division module: obtain public surveillance video clips containing multiple target types and behavior patterns from the Internet through the API interface, annotate the surveillance clips, obtain target recognition and behavior analysis data sets, and use 80% of the data sets as the first data set and 20% as the second data set; Data input module: inputting the annotated first data set into the deep neural network architecture; training the target recognition model through the deep neural network architecture; Model building module: building a behavior analysis model suitable for behavior analysis, and training the behavior analysis model through the first data set; Fusion optimization module: Fusion and optimization of the target recognition model and the behavior analysis model. Through joint training and multi-task learning, some underlying features of the target recognition model and the behavior analysis model are shared and their respective results are output. Performance evaluation module: evaluates the performance indicators of the trained monitoring model through the second data set, and adjusts and optimizes the model according to the evaluation results; Model conversion module: converts the adjusted and optimized monitoring model and deploys the converted monitoring model in the front-end intelligent monitoring device. 8. The intelligent monitoring and management system according to claim 6, characterized in that the data transmission module comprises: Data acquisition module: The front-end intelligent monitoring device acquires the video stream through the built-in camera, uses the acquired video stream as the first monitoring data, pre-processes the first monitoring data, and divides the continuous video stream into single-frame images; Target detection module: Each frame of the image is input into the deployed target recognition model, which performs target detection in real time based on the trained parameters and outputs target detection information. The main task of this stage is to accurately identify and extract the target object of focus from the complex background. Behavior judgment module: inputs a continuous multi-frame image sequence into the behavior analysis model, and judges whether the target has abnormal behavior or specific behavior patterns by analyzing the movement trajectory and morphological changes of the target object in the time dimension; Saving module: according to the results of target detection and behavior analysis, the first analysis result is generated by summarizing, and the frames where the behavior events occur or have key information are selected and saved as key frames; Encryption and compression module: encrypt and compress the first analysis result and the selected key frame, and transmit the encrypted and compressed first analysis result and the selected key frame to the edge computing node through a multi-channel transmission protocol. 9. The intelligent monitoring and management system according to claim 6, characterized in that the data uploading module comprises: Decryption and decompression module: The edge computing node receives the first analysis result and key frame data sent by the front-end intelligent monitoring device, and performs decryption and decompression; Preliminary screening module: Preliminary screening of target information and behavior events in the first analysis result according to preset rules, eliminating false positives and data that do not meet the conditions, and verifying whether the content of the key frame matches the analysis result; Secondary optimization module: The filtered data is analyzed and optimized again through parallel processing algorithms, and abnormal events are identified and marked from the optimized data according to preset business rules and security policies, and information of key time and space points is extracted to obtain the second monitoring data; Integration module: If there are multiple monitoring sources, the edge computing node is used to fuse and correlate cross-source data, and integrate related events or information from different sources; Priority sorting module: compress and encrypt the second monitoring data, prioritize the packaged second monitoring data according to the severity of the abnormal event or the urgency level of important information, and transmit the compressed and encrypted second monitoring data to the cloud server according to the priority through a multi-channel transmission protocol. 10. The intelligent monitoring and management system according to claim 6, characterized in that the result feedback module comprises: Space allocation module: the cloud server receives the second monitoring data uploaded by the edge computing node, decrypts and decompresses the second monitoring data, and stores the second monitoring data in different storage spaces according to the priority order; Preprocessing module: the cloud server re-preprocesses the second monitoring data stored in different storage spaces, and performs feature engineering processing on the re-preprocessed second monitoring data; Analysis result module: The pre-processed data is analyzed in depth through machine learning algorithms to obtain the second analysis result; and the third analysis result is obtained based on spatiotemporal sequence analysis, cluster analysis and association rule analysis algorithms; The third analysis module: If it is a specific scenario, targeted analysis is performed through domain knowledge and expert systems, and the fourth analysis result is obtained. Fourth analysis module: Based on various analysis results, a comprehensive situation model is constructed, and the regional situation of the comprehensive situation model is evaluated to obtain the fourth analysis result; Report feedback module: summarize the first to fourth analysis results and generate a comprehensive situation report, which is pushed to relevant managers in various ways; relevant managers take corresponding actions after receiving the comprehensive situation report.