CN103093212B - The method and apparatus of facial image is intercepted based on Face detection and tracking - Google Patents

Abstract

The invention discloses a method and a device for intercepting a face image based on face detection and tracking, and belongs to the technical field of face tracking. The method includes: using a cascade classifier to detect the face of the image to be detected; when the face target is detected, using the mean value tracking algorithm to perform face tracking on the face target; when the face target leaves the detection area, in each On the frame, judge whether the face detection and face tracking in the same frame correspond to the same face target according to the position of the target, and select each frame in which the face detection and face tracking correspond to the same face target; in each selected frame , calculate the coincidence degree between the face detection window and the face tracking window in the same frame, and use the face image obtained by face detection on the frame with the largest coincidence degree as the intercepted face image. The device includes: a detection module, a tracking module, a judging module and an intercepting module. The invention intercepts a relatively clear human face image, and improves the accuracy and tracking effect of human face tracking.

Description

Method and device for capturing face images based on face detection and tracking

technical field

The present invention relates to the technical field of human face tracking, in particular to a method and device for capturing human face images based on human face detection and tracking.

Background technique

With the improvement of security requirements, the commercial value of people flow statistics, personnel feature recognition, and face recognition technology has begun to be revealed and gradually started to be applied. As an important part of these tasks, face detection and face tracking have a very important role and significance. In recent years, researchers have invested a lot of time and energy in this field, devoting themselves to developing fast and accurate face detection methods and tracking methods.

Face detection refers to the process of determining the position and size of a human face in a given picture. At present, the commonly used face detection method is a face detection method based on Haar features and Boosted cascade. The core idea of the algorithm is to iteratively select multiple weak classifiers with different classification capabilities to form a strong classifier, and then combine multiple strong classifiers sequentially to form a cascade classifier as the final face detection device.

Face tracking refers to the process of determining the trajectory and size change of a certain face in the input image sequence. Face tracking technology has important potential application value. As a key technology in the fields of automatic face recognition, video retrieval, video surveillance, etc., it is generally valued by researchers. At present, commonly used face tracking methods include mean-shift algorithm, cam-shift algorithm, particle filter and so on.

However, under normal circumstances, the size, shape, illumination and other conditions of the target will change during the moving process of the target tracked by the face. With the increase of the number of tracking frames in the current face tracking technology, the tracking error will gradually increase, resulting in poor tracking effect and low accuracy of the tracking results. When the face appears in the video, a clearer frame is saved. It is also an important requirement and function in the security system to use it as a database in the future.

Contents of the invention

In order to improve the accuracy of face tracking results, the present invention provides a method and device based on the method of face detection and tracking to intercept a face image. Described technical scheme is as follows:

On the one hand, the present invention provides a kind of method based on face detection and tracking interception face image, described method comprises:

Use cascaded classifiers to perform face detection on the image to be detected;

When a face target is detected, convert the current frame to be tracked into an HSV image, and obtain information on its brightness, chroma, and saturation; randomly select the position of the face target around the position of the previous frame as the center point , extract the image area information with the target area as the size; then according to the position information of the face target in the last frame and the initial position information of the face target, use a predetermined state equation to estimate the position of the face target in the current The position where the frame to be tracked appears; calculate the color information in the HSV space, and calculate the weight of each estimated position according to the color similarity, assign a high weight to the value with a large similarity, and assign a low weight to a value with a small similarity. Obtaining a weight average value for all the calculated weights, obtaining a corresponding position according to the weight average value, and performing face tracking on the human face target according to the obtained position;

When the human face target leaves the detection area, on each frame of human face detection and human face tracking, calculate the upper left corner vertex of the target position of human face detection in this frame and the upper left corner vertex of the target position of human face tracking and calculate the ratio of the face detection window side length to the face tracking window side length in the frame, when the distance is less than or equal to the preset distance threshold and the ratio is less than or equal to the preset ratio threshold, Determine that the face detection and face tracking in the frame correspond to the same face target, and select each frame in which the face detection and face tracking correspond to the same face target;

In each selected frame, calculate the coincidence degree between the window of face detection and the window of face tracking in the same frame, compare all the coincidence degrees obtained by calculation, and compare the face image obtained by face detection on the frame with the largest coincidence degree as the intercepted face image.

Wherein, the method also includes:

When it is judged according to the position of the target that the face detection and face tracking in a certain frame correspond to the same face target, replace the face image obtained by face tracking in the frame with the face image obtained by face detection to replace The final face image continues to face tracking.

Wherein, the method also includes:

When it is judged according to the position of the target that the face detection and face tracking in a certain frame correspond to different face targets, the face image detected by the face in the frame is used as a new face target, and the new face target is used for the new target. The face target starts face tracking.

On the other hand, the present invention also provides a kind of device based on face detection and tracking interception face image, and described device comprises:

Detection module, for adopting cascade classifier to carry out face detection on the image to be detected;

The tracking module is used to convert the current frame to be tracked into an HSV image when a human face target is detected, and obtain information on its brightness, chroma, and saturation; randomly select the position of the human face target around the previous frame The position of the face is used as the center point to extract the image area information with the target area as the size; then according to the position information of the face target in the last frame and the initial position information of the face target, use a predetermined state equation to estimate the face The position where the target appears in the current frame to be tracked; the color information is calculated in the HSV space, and the weight of each estimated position is calculated according to the color similarity, and the value with a large similarity is assigned a high weight, and the value with a small similarity is assigned a high weight Giving a low weight, obtaining a weight average value for all calculated weights, obtaining a corresponding position according to the weight average value, and performing face tracking on the human face target according to the obtained position;

Judgment module, for when the face target leaves the detection area, on each frame of face detection and face tracking, calculate the upper left corner vertex of the target position of face detection in the frame and the target of face tracking The distance between the vertices in the upper left corner of the position, and calculate the ratio of the face detection window side length to the face tracking window side length in the frame, when the distance is less than or equal to the preset distance threshold and the ratio is less than or equal to the preset When the ratio threshold value of , it is judged that the face detection in the frame corresponds to the same face target with the face tracking, and each frame of the same face target corresponding to the face detection and the face tracking is selected;

The interception module is used to calculate the coincidence degree between the window of face detection and the window of face tracking in the same frame in each selected frame, compare all the coincidence degrees obtained by calculation, and detect the face on the frame with the largest coincidence degree The obtained face image is used as the intercepted face image.

Wherein, the tracking module is also used for:

When the judging module judges according to the position of the target that the face detection and face tracking in a certain frame correspond to the same face target, replace the face image obtained by face tracking in the frame with the face obtained by face detection Image, continue face tracking with the replaced face image.

Wherein, the tracking module is also used for:

When the judging module judges according to the position of the target that the face detection and face tracking in a certain frame correspond to different face targets, the face image detected by the face in the frame is used as a new face target, Start face tracking for the new face target.

The beneficial effect brought by the technical solution provided by the present invention is: the face target detected by the cascade classifier is tracked by the mean value tracking algorithm. On each frame of the face detection and face tracking, select each frame corresponding to the same face target; in each selected frame, calculate the coincidence degree of the window of the face detection and the window of the face tracking in the same frame, The face image obtained by face detection on the frame with the largest overlap is used as the intercepted face image. On the basis of making full use of detection resources and tracking resources, a clearer face image is intercepted, which improves the accuracy of face tracking , which improves the tracking effect and provides data support for capturing clear face images.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

Fig. 1 is the flow chart of the method for intercepting a face image based on face detection and tracking provided by an embodiment of the present invention;

Fig. 2 is a schematic diagram of the detection of the cascade classifier provided by the embodiment of the present invention;

Fig. 3 is the flow chart of the method for intercepting a face image based on face detection and tracking provided by another embodiment of the present invention;

Fig. 4 is a schematic diagram of calculating coincidence degree provided by an embodiment of the present invention;

FIG. 5 is a structural diagram of a device for capturing and capturing a face image based on face detection and tracking provided by an embodiment of the present invention;

Fig. 6 is a structural diagram of a device for capturing and capturing a face image based on face detection and tracking provided by another embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the implementation manner of the present invention will be further described in detail below in conjunction with the accompanying drawings.

Embodiments of the invention relate to cascaded classifiers. The cascade classifier is composed of multiple classifiers in series, and the number of the multiple strong classifiers is also called the number of stages of the cascade classifier. For example, a 10-level cascade of classifiers consists of 10 strong classifiers and so on. Usually, the cascade classifier is trained for face detection, and can be trained with pre-prepared positive samples and negative samples. Wherein, the size and quantity of the positive sample and the negative sample are not specifically limited in the present invention. For example, the positive sample is a face image with a size of 20*20 pixels, the sample size is 10,000, and the negative sample is any different image in nature. A 20*20 pixel image containing a face, with a sample size of 20,000 and so on. Preferably, the images used as samples should be as diverse as possible, and it is more appropriate to draw materials from various living environments of people. After the training is completed, the classifier can be adjusted according to whether the training result reaches the predetermined detection rate, false alarm rate and other indicators, such as increasing the number of cascaded classifiers and so on. Wherein, the numerical value of predetermined detection rate, false alarm rate the present invention also does not limit to this, can set according to need, as presetting the face detection rate of each level classifier 99%, the false alarm rate of non-human face The police rate is 30% and so on. For each level of classifier, during the training process, for the detected negative samples that are identified as non-face negative samples, before entering the next level of classifier, it is necessary to perform sample replacement. The detected negative samples are replaced with other negative samples of the same number, and then input to the next classifier together with the negative samples that are not detected as non-human faces to continue training.

Among them, the cascaded classifiers perform face detection step by step. Starting from the first-level classifier, the results of face detection by each classifier are input to the next-level classifier to continue face detection. Until the last level of classifier is detected, the result of face detection is output.

Referring to Fig. 1, an embodiment of the present invention provides a kind of method based on face detection and tracking interception face image, comprising:

101: Use a cascade classifier to perform face detection on the image to be detected.

102: When a face target is detected, perform face tracking on the face target using a mean value tracking algorithm.

103: When the face target leaves the detection area, on each frame of face detection and face tracking, judge whether the face detection and face tracking in the same frame correspond to the same face target according to the position of the target, and select Face detection and face tracking correspond to each frame of the same face target.

104: In each selected frame, calculate the coincidence degree of the window of face detection and the window of face tracking in the same frame, compare all the coincidence degrees obtained by calculation, and calculate the person detected by face detection on the frame with the largest degree of coincidence The face image is used as the intercepted face image.

The image to be detected is usually a video image, which is a continuous video, and of course it can also be a group of static images, which is not specifically limited in the present invention. When the cascade classifier detects the image to be detected, it can perform face detection according to the preset detection window, and detect whether there is a human face in each detection window. The size of the detection window is not specifically limited in the present invention, and can be set as required, such as a window with a size of 20*20 pixels, a window with a size of 25*25 pixels, or a window with a size of 30*30 pixels, etc. The translation sequence of the detection window on the image may be from top to bottom and from left to right, which is not specifically limited in the present invention.

Referring to FIG. 2 , it is a schematic diagram of face detection performed by cascade classifiers provided in this embodiment. Among them, the cascade classifier is N-level, and there are N classifiers in total. Starting from classifier 1, face detection is performed on the image to be detected step by step, and the detected face is passed through the classifier as the output result to enter the next classifier for further detection, and the detected non-face is output as the result of rejection to the non-human face pool. After the last classifier N is detected, the final result is the detected face target.

The mean value tracking algorithm refers to taking the mean value of the weights of the estimated position of the human face target, and tracking according to the position corresponding to the mean value. Specifically, when a human face target is detected, using the mean value tracking algorithm to perform face tracking on the human face target may include:

When a human face target is detected, estimate the position where the human face target appears in the next frame;

Calculate the weight of each estimated position, and calculate the weight average of all the calculated weights;

The corresponding position is obtained according to the weight average value, and the face tracking is performed on the face target according to the obtained position.

In this embodiment, on each frame of face detection and face tracking, it is judged according to the position of the target whether the face detection and face tracking in the same frame correspond to the same face target, including:

On each frame of face detection and face tracking, calculate the distance between the upper left corner vertex of the target position of face detection in the frame and the upper left corner vertex of the target position of face tracking, and calculate the distance between the target position of the face in the frame The ratio of the side length of the face detection window to the side length of the face tracking window. When the distance is less than or equal to the preset distance threshold and the ratio is less than or equal to the preset ratio threshold, it is determined that the face detection and human face in the frame Tracking corresponds to the same face target.

In this embodiment, the method also includes:

When it is judged according to the position of the target that the face detection and face tracking in a certain frame correspond to the same face target, replace the face image obtained by face tracking in the frame with the face image obtained by face detection to replace The final face image continues to face tracking.

In this embodiment, the method also includes:

When it is judged according to the position of the target that the face detection and face tracking in a certain frame correspond to different face targets, the face image detected by the face in the frame is used as a new face target, and the new face target is used for the new target. The face target starts face tracking.

In this embodiment, face detection is performed in real time, and detection may be performed every frame, or every few frames, such as face detection every two frames, etc., which is not specifically limited in the present invention. The step size of face tracking and the step size of face detection can be the same or different. Preferably, in this embodiment, face detection and face tracking adopt the same step size. For example, face detection is performed in each frame. and tracking or perform face detection and tracking every two frames and so on.

In the above-mentioned method provided by this embodiment, the face target detected by the cascade classifier is used for face tracking through the mean tracking algorithm. When the face target leaves the detection area, on each frame of face detection and face tracking, Select each frame corresponding to the same face target for face detection and face tracking; in each selected frame, calculate the coincidence degree between the window of face detection and the window of face tracking in the same frame, and the window with the largest coincidence degree The face image obtained by face detection on the frame is used as the intercepted face image. On the basis of making full use of detection resources and tracking resources, a clearer face image is intercepted, which improves the accuracy of face tracking and improves the tracking effect. , and provides data support for intercepting clear face images.

Referring to Fig. 3, another embodiment of the present invention provides a kind of method based on face detection and tracking interception face image, comprising:

301: Perform face detection on the image to be detected by using cascaded classifiers.

The image to be detected is usually a video image, which is a continuous video, and of course it can also be a group of static images, which is not specifically limited in the present invention.

When the cascade classifier detects the image to be detected, it can perform face detection according to the preset detection window, and detect whether there is a human face in each detection window. The size of the detection window is not specifically limited in the present invention, and can be set as required, such as a window with a size of 20*20 pixels, a window with a size of 25*25 pixels, or a window with a size of 30*30 pixels, etc. The translation sequence of the detection window on the image may be from top to bottom and from left to right, which is not specifically limited in the present invention.

302: When a face target is detected, perform face tracking on the face target using a mean value tracking algorithm.

In this embodiment, face tracking is started when a face target is detected. The position of face tracking is estimated according to the algorithm. Usually, multiple positions are estimated, and the multiple positions represent possible positions of the face target in the next frame. According to the weight corresponding to each position, a suitable position can be calculated from the multiple positions for tracking.

This step may specifically include the following steps:

When a human face target is detected, estimate the position where the human face target appears in the next frame;

Calculate the weight of each estimated position, and calculate the weight average of all the calculated weights;

The corresponding position is obtained according to the weight average value, and the face tracking is performed on the face target according to the obtained position.

Wherein, a predetermined state equation, such as a second-order Markov chain autoregressive equation, may be used to estimate the position where the face target appears in the next frame. The calculating weight may calculate color information in HSV space, and calculate the weight of each estimated position according to color similarity. Specifically, the current frame to be tracked can be converted into an HSV image, and its brightness, chroma, and saturation information can be obtained; then the position of the face target around the position of the previous frame is randomly selected as the center point, and the target area is extracted is the size of the image area information; then according to the position information of the face target in the previous frame and the initial position information of the face target, use a suitable state equation such as the second-order Markov chain autoregressive equation to estimate the face target in this frame possible location. The center point selected each time is different, and the estimated position in this frame is also different. For example, select 300 different center points to obtain 300 estimated positions in total, calculate the feature information of the HSV space of each position, use the histogram to compare with the initial image, and obtain the similarity data. Values with high weights are given high weights, values with low similarity are given low weights, and the final location center is calculated by the weighted average of all estimated locations.

In this embodiment, there may be one estimated position of the human face target appearing in the next frame, and usually multiple positions. Each estimated position will correspond to a weight, which represents the possibility of the position, the greater the weight, the greater the possibility of the face target appearing at the position, and the smaller the weight, the face target It is less likely to appear in this position. When multiple positions are estimated, the weight of each position is obtained to obtain multiple weights, and then the multiple weights are averaged to obtain a weight average.

303: When the face target leaves the detection area, on each frame of face detection and face tracking, calculate the upper left corner vertex of the target position of face detection in the frame and the upper left corner of the target position of face tracking The distance between the vertices, and calculate the ratio of the face detection window side length to the face tracking window side length in the frame, when the distance is less than or equal to the preset distance threshold and the ratio is less than or equal to the preset ratio threshold , it is determined that the face detection and face tracking in the frame correspond to the same face target.

Wherein, the distance threshold and the ratio threshold can be preset according to needs, and the present invention does not limit specific values. For example, the distance threshold can be set to 5% of the side length of the face detection window, and the ratio threshold can be set to 5%, 10%, and so on.

In this embodiment, the distance between the upper left corner vertex of the target position of face detection and the upper left corner vertex of the target position of face tracking is less than or equal to the distance threshold, and the side length of the face detection window is equal to the side length of the face tracking window. When the ratio is less than or equal to the ratio threshold, it is considered that the two positions of face detection and face tracking are very close, and these two positions can be regarded as the position of the same face target, so that it can be determined that the detected face target is also currently being tracked. Tracked face targets.

304: Select each frame corresponding to the same face target for face detection and face tracking.

305: In each selected frame, calculate the coincidence degree of the window of face detection and the window of face tracking in the same frame, compare all the coincidence degrees obtained by calculation, and calculate the human face detected by the frame with the largest degree of coincidence The face image is used as the intercepted face image.

Referring to Fig. 4, in the present embodiment, the coincidence degree of the window of face detection and the window of face tracking in any frame can be calculated in the following manner:

Among them, the vertex coordinates of the upper left corner of the window for face detection are (x ₁ , y ₁ ), and the side length is d ₁ , and the vertex coordinates of the upper left corner of the window for face tracking are (x ₂ , y ₂ ), and the side length is d ₂ , here it is assumed that both windows are square.

First, calculate the width of the overlapping window: compare the sizes of x ₁ and x ₂ , and take the value x_maxleft=max(x ₁ , x ₂ ) as the abscissa of the top left corner of the overlapping window; then calculate x ₁ +d ₁ and x ₂ +d ₂ , take the value x_minright=min(x ₁ +d ₁ ,x ₂ +d ₂ ) as the abscissa of the upper right vertex of the overlapping window; at this time, the width of the overlapping window can be calculated as w=x_minright-x_maxleft .

Then, calculate the height of the coincident window: compare the sizes of y ₁ and y ₂ , and take the value y_maxtop=max(y ₁ , y ₂ ) as the ordinate of the top left corner of the coincident window; then calculate y ₁ +d ₁ and y ₂ +d ₂ , take the value y_minbottom=min(y ₁ +d ₁ ,y ₂ +d ₂ ) as the ordinate of the bottom left corner vertex of the overlapping window; at this time, the height of the overlapping window can be calculated as h=y_minbottom-y_maxtop .

Finally, according to the width and height of the overlapping window, the area of the overlapping window is calculated as S=w×h, and the area of the face detection window is calculated According to the area of the coincident window and the area of the face detection window, the coincidence degree of the face detection window and the face tracking window is calculated as destination=min(S, S_detect)/S*100%.

In this embodiment, the largest coincidence degree is selected among the calculated coincidence degrees, and the face image of the intra-frame face detection corresponding to the maximum coincidence degree is used as the final intercepted face image, and the clearest result can be obtained . If there are multiple maximum coincidence degrees, the first frame that appears in the multiple frames corresponding to the multiple maximum coincidence degrees can be selected, and the face image detected by the face in the frame is used as the last intercepted face image .

In the above steps, the tracking of the face target has been started. If the cascade classifier detects multiple face targets, the tracking of the multiple face targets will be started, that is, each detected face target is tracked separately. Here Do not explain one by one.

Further, the above method may also include the following steps:

When it is judged according to the position of the target that the face detection and face tracking in a certain frame correspond to the same face target, replace the face image obtained by face tracking in the frame with the face image obtained by face detection to replace The final face image continues to face tracking.

And/or, the above method may also include the following steps:

When it is judged according to the position of the target that the face detection and face tracking in a certain frame correspond to different face targets, the face image detected by the face in the frame is used as a new face target, and the new face target is used for the new target. The face target starts face tracking.

In this embodiment, the position obtained by tracking the human face target in the current frame can be called a tracking position (English: TrackLocation). location, which may be referred to as a detection location (English: DetectLocation). Obviously, TrackLocation and DetectLocation are two results obtained by face detection and face tracking for the same face target, and these two results are necessarily related. In actual situations, these two positions are usually very close, but in terms of accuracy, DetectLocation is generally more accurate in judging the target position, while TrackLocation will gradually increase the error as the number of tracking frames of the target increases. This is because as the target moves, the size and shape of the target, lighting and other conditions are changing, and the gap with the initial information is getting bigger and bigger, and the tracking effect will be getting worse. Therefore, in this embodiment, the position of the face image obtained by tracking is replaced by the position of the face image obtained by face detection, and TrackLocation=DetectLocation is set to continue tracking with the replaced position, which can improve tracking accuracy and reduce tracking errors. When the tracked face target leaves the tracking area, the face tracking result is compared with the face detection result, and the frame with the closest position between the face tracking result and the face detection result is found during the tracking process. The face has strong trackability, less interference, and high definition. Therefore, the face image obtained by face detection on this frame is selected as the final screenshot, which has high accuracy.

In this embodiment, when the tracked face target moves out of the tracking area, the tracking can be stopped, and all tracking resources occupied by the face target can be released, so that tracking resources can be saved and waste of resources can be avoided.

In the above-mentioned method provided by this embodiment, the face target detected by the cascade classifier is used for face tracking through the mean tracking algorithm. When the face target leaves the detection area, on each frame of face detection and face tracking, Select each frame corresponding to the same face target for face detection and face tracking; in each selected frame, calculate the coincidence degree between the window of face detection and the window of face tracking in the same frame, and the window with the largest coincidence degree The face image obtained by face detection on the frame is used as the intercepted face image. On the basis of making full use of detection resources and tracking resources, a clearer face image is intercepted, which improves the accuracy of face tracking and improves the tracking effect. , and provides data support for intercepting clear face images. In this embodiment, the mean value tracking algorithm is used to track, and the weights of each estimated position are averaged. Since each estimated position reflects the various possibilities of the face target appearing in the next frame, the position corresponding to the weighted average value is more accurate. It fully and more accurately represents the position of the face target in the next frame, and the accuracy of face tracking is higher according to the position corresponding to the weight average. Comparing the tracking results with the detection results and verifying each other provides a clear data basis for the process of capturing clear face images. When the tracking result is very close to the detection result, it can be considered that the face at this time has strong traceability, less interference, and high definition, so this frame is selected as the final screenshot, which is also the goal of the present invention and current market products. The biggest difference.

Referring to Fig. 5, an embodiment of the present invention provides a device for capturing a face image based on face detection and tracking, including:

The detection module 501 is used to perform face detection on the image to be detected by using a cascade classifier;

Tracking module 502, for when detection module 501 detects human face target, use mean value tracking algorithm to carry out human face tracking to described human face target;

Judging module 503, for when the face target leaves the detection area, on each frame of face detection and face tracking, judge whether the face detection and face tracking in the same frame correspond to the same person according to the position of the target Face target, select each frame corresponding to the same face target for face detection and face tracking;

The intercepting module 504 is used to calculate the coincidence degree of the window of face detection and the window of face tracking in the same frame in each selected frame, compare all the coincidence degrees obtained by the calculation, and place the face on the frame with the largest degree of coincidence The detected face image is used as the intercepted face image.

Referring to Figure 6, the tracking module 502 includes:

An estimating unit 502a, configured to estimate the position where the human-face object appears in the next frame when the detection module detects the human-face object;

Calculation unit 502b, configured to calculate the weight of each estimated position, and calculate the weight average of all the calculated weights;

The tracking unit 502c is configured to acquire a corresponding position according to the weight average value, and perform face tracking on the face target according to the acquired position.

Wherein, the judging module 503 is used for:

On each frame of face detection and face tracking, calculate the distance between the upper left corner vertex of the target position of face detection in the frame and the upper left corner vertex of the target position of face tracking, and calculate the distance between the target position of the face in the frame The ratio of the side length of the face detection window to the side length of the face tracking window. When the distance is less than or equal to the preset distance threshold and the ratio is less than or equal to the preset ratio threshold, it is determined that the face detection and human face in the frame Tracking corresponds to the same face target.

Wherein, the tracking module 502 is also used for:

When the judging module 503 judges according to the position of the target that the face detection and face tracking in a certain frame correspond to the same face target, replace the face image obtained by face tracking in the frame with the face image obtained by face detection , continue face tracking with the replaced face image.

Wherein, the tracking module 502 is also used for:

When the judging module 503 judges according to the position of the target that the face detection in a certain frame is different from the face tracking corresponding to the face target, the face image detected by the face in the frame is used as a new face target. The new face object starts face tracking.

The above-mentioned device provided in this embodiment can execute the method provided in any one of the above-mentioned method embodiments. For the detailed process, refer to the description in the method embodiment, and details are not repeated here. The device can be applied to electronic equipment such as computers, which is not specifically limited in the present invention.

The above-mentioned device provided in this embodiment performs face tracking on the face target detected by the cascade classifier through the mean value tracking algorithm. When the face target leaves the detection area, on each frame of face detection and face tracking, Select each frame corresponding to the same face target for face detection and face tracking; in each selected frame, calculate the coincidence degree between the window of face detection and the window of face tracking in the same frame, and the window with the largest coincidence degree The face image obtained by face detection on the frame is used as the intercepted face image. On the basis of making full use of detection resources and tracking resources, a clearer face image is intercepted, which improves the accuracy of face tracking and improves the tracking effect. , and provides data support for intercepting clear face images. The average tracking algorithm is used for tracking, and the weight of each estimated position is averaged. Since each estimated position reflects the various possibilities of the face target appearing in the next frame, the position corresponding to the weighted average is more comprehensive and accurate. The ground represents the position where the face target appears in the next frame, and the accuracy of face tracking is higher according to the position corresponding to the weight average. Usually, the accuracy of the face detection result is usually higher than the accuracy of the face tracking result, therefore, replace the position of the face image obtained by tracking with the position of the face image obtained by face detection, and continue with the replaced position Tracking can improve tracking accuracy, reduce tracking errors, and provide data basis for screenshots of clear face images.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above embodiments can be completed by hardware, and can also be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium. The above-mentioned The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, and the like.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (6)

1. a kind of method based on face detection and tracking interception face image, it is characterized in that, described method comprises: Use cascaded classifiers to perform face detection on the image to be detected; When a face target is detected, convert the current frame to be tracked into an HSV image, and obtain information on its brightness, chroma, and saturation; randomly select the position of the face target around the position of the previous frame as the center point , extract the image area information with the target area as the size; then according to the position information of the face target in the last frame and the initial position information of the face target, use a predetermined state equation to estimate the position of the face target in the current The position where the frame to be tracked appears; calculate the color information in the HSV space, and calculate the weight of each estimated position according to the color similarity, assign a high weight to the value with a large similarity, and assign a low weight to a value with a small similarity. Obtaining a weight average value for all the calculated weights, obtaining a corresponding position according to the weight average value, and performing face tracking on the human face target according to the obtained position; When the human face target leaves the detection area, on each frame of human face detection and human face tracking, calculate the upper left corner vertex of the target position of human face detection in this frame and the upper left corner vertex of the target position of human face tracking and calculate the ratio of the face detection window side length to the face tracking window side length in the frame, when the distance is less than or equal to the preset distance threshold and the ratio is less than or equal to the preset ratio threshold, Determine that the face detection and face tracking in the frame correspond to the same face target, and select each frame in which the face detection and face tracking correspond to the same face target; In each selected frame, calculate the coincidence degree between the window of face detection and the window of face tracking in the same frame, compare all the coincidence degrees obtained by calculation, and compare the face image obtained by face detection on the frame with the largest coincidence degree as the intercepted face image. 2. The method according to claim 1, characterized in that the method further comprises: When it is judged according to the position of the target that the face detection and face tracking in a certain frame correspond to the same face target, replace the face image obtained by face tracking in the frame with the face image obtained by face detection to replace The final face image continues to face tracking. 3. The method according to claim 1, characterized in that the method further comprises: When it is judged according to the position of the target that the face detection and face tracking in a certain frame correspond to different face targets, the face image detected by the face in the frame is used as a new face target, and the new face target is used for the new target. The face target starts face tracking. 4. a kind of device based on face detection and tracking interception face image, it is characterized in that, described device comprises: Detection module, for adopting cascade classifier to carry out face detection on the image to be detected; The tracking module is used to convert the current frame to be tracked into an HSV image when a human face target is detected, and obtain information on its brightness, chroma, and saturation; randomly select the position of the human face target around the previous frame The position of the face is used as the center point to extract the image area information with the target area as the size; then according to the position information of the face target in the last frame and the initial position information of the face target, use a predetermined state equation to estimate the face The position where the target appears in the current frame to be tracked; the color information is calculated in the HSV space, and the weight of each estimated position is calculated according to the color similarity, and the value with a large similarity is assigned a high weight, and the value with a small similarity is given a high weight Giving a low weight, obtaining a weight average value for all calculated weights, obtaining a corresponding position according to the weight average value, and performing face tracking on the human face target according to the obtained position; Judgment module, for when the face target leaves the detection area, on each frame of face detection and face tracking, calculate the upper left corner vertex of the target position of face detection in the frame and the target of face tracking The distance between the upper left corner vertices of the position, and calculate the ratio of the face detection window side length to the face tracking window side length in the frame, when the distance is less than or equal to the preset distance threshold and the ratio is less than or equal to the preset When the ratio threshold value of , it is judged that the face detection in the frame corresponds to the same face target with the face tracking, and each frame of the same face target corresponding to the face detection and the face tracking is selected; The interception module is used to calculate the coincidence degree between the window of face detection and the window of face tracking in the same frame in each selected frame, compare all the coincidence degrees obtained by calculation, and detect the face on the frame with the largest coincidence degree The obtained face image is used as the intercepted face image. 5. The device according to claim 4, wherein the tracking module is also used for: When the judging module judges according to the position of the target that the face detection and face tracking in a certain frame correspond to the same face target, replace the face image obtained by face tracking in the frame with the face obtained by face detection Image, continue face tracking with the replaced face image. 6. The device according to claim 4, wherein the tracking module is also used for: When the judging module judges according to the position of the target that the face detection and face tracking in a certain frame correspond to different face targets, the face image detected by the face in the frame is used as a new face target, Start face tracking for the new face target.