CN107229906A - A kind of automobile overtaking's method for early warning based on units of variance model algorithm - Google Patents

Abstract

The invention relates to a vehicle overtaking early warning method based on a variable component model algorithm, comprising the following steps: the rear vehicle overtaking early warning step, the vehicle uses a rear view camera combined with a variable component model algorithm to identify the rear vehicle, and through identifying the rear vehicle and its own The positional relationship between the car and the lane line determines whether the vehicle behind is overtaking and issues an early warning to the driver; in the overtaking warning step of the vehicle in front, the car recognizes the vehicle in front through the front-view camera combined with the variable component model algorithm, and recognizes the vehicle in front Based on the positional relationship with the vehicle and the lane line, it is judged whether the vehicle needs to overtake and an early warning is given to the driver when overtaking is dangerous. Compared with the prior art, the present invention has the advantages of accurate early warning, high early warning speed, convenient realization and the like.

Description

A Vehicle Overtaking Early Warning Method Based on Variable Component Model Algorithm

technical field

The invention relates to the field of overtaking early warning, in particular to an automobile overtaking early warning method based on a variable component model algorithm.

Background technique

Overtaking is one of the most common driving behaviors for drivers. According to statistics, when a driver travels a distance of 100km at 90km/h on the expressway, he will perform about 50 overtaking behaviors on the way. In recent years, traffic accidents caused by improper overtaking in our country have shown an obvious upward trend, especially on expressways, where more than 60% of traffic accidents are related to overtaking. When implementing overtaking, the driver must adjust the driving strategy in real time to achieve overtaking behavior according to the current vehicle speed, vehicle distance, traffic flow status, and road traffic facilities and other surrounding environmental information. Avoiding vehicle collisions caused by overtaking can be achieved by controlling the relative speed between vehicles and increasing the longitudinal distance between vehicles.

The patent document with the publication number CN105216797A discloses a technical solution called the overtaking method and system, which includes a detection module and a processing module, and obtains the speed and position information of the obstacle through the camera and the front radar to determine whether the target vehicle can perform Overtaking action; the patent document whose publication number is CN101326511 discloses a technical solution titled as a method for detecting or predicting vehicle overtaking, in an embodiment of the solution, wireless communication is used to read the navigation systems of other vehicles to determine the 1. Whether the vehicle will overtake.

At present, in domestic technologies related to overtaking and lane changing, most of them use radar or short-distance information communication systems to prevent traffic accidents during overtaking. This solution is costly due to the use of multiple sensors and cannot be implemented in real-time scenarios. Quickly and effectively identify vehicles and lane line information, or just predict the distance information between obstacles and vehicles.

Contents of the invention

The object of the present invention is to provide a kind of vehicle overtaking early warning method based on variable component model algorithm for above-mentioned problem.

The purpose of the present invention can be achieved through the following technical solutions:

A vehicle overtaking warning method based on a variable component model algorithm, said method comprising the following steps:

In the overtaking warning step of the rear vehicle, the car recognizes the rear vehicle through the rear view camera combined with the variable component model algorithm, and by identifying the positional relationship between the rear vehicle, the vehicle and the lane line, judges whether the rear vehicle is overtaking and sends a signal to the driver. early warning;

In the overtaking warning step of the vehicle in front, the car recognizes the vehicle in front through the front-view camera combined with the variable component model algorithm. By identifying the positional relationship between the vehicle in front, the vehicle and the lane line, it is judged whether the vehicle needs to overtake and whether the overtaking is necessary. Alerts the driver in case of danger.

The steps of the vehicle overtaking warning in the rear are specifically:

A1) The car detects whether other vehicles appear in the image captured by the rear-view camera according to the variable component model algorithm, and if so, enters step A2), otherwise continues to detect;

A2) The car judges whether the distance between the detected vehicle and the vehicle decreases, if so, enter step A3), otherwise return to step A1);

A3) The car recognizes the lane line, and judges whether the car and the detected vehicle are located in the same lane line, and if so, sends an early warning to the driver, otherwise proceeds to step A4);

A4) After the detected vehicle disappears from the rear view camera, the car enters the overtaking warning step of the vehicle in front.

The overtaking warning steps of the vehicle in front are specifically as follows:

B1) The car detects whether other vehicles appear in the image taken by the front-view camera according to the variable component model algorithm, if so, enters step B2), otherwise continues to detect;

B2) the car judges whether the distance between the detected vehicle and the vehicle decreases, if so, enter step B3), otherwise return to step B1);

B3) The car recognizes the lane line, and judges whether the car and the detected vehicle are located in the same lane line, and if so, sends an early warning to the driver, otherwise proceeds to step B4);

B4) The car enters the rear vehicle overtaking warning step after the detected vehicle disappears from the front view camera.

The algorithm of the variable component model is specifically:

A11) Build a scale pyramid according to the photographed images of the rear-view camera;

A12) In each scale layer in the scale pyramid, the image is matched with the model trained by the variable component model algorithm by the sliding window detection method, and the matching score is calculated;

A13) Select the scale layer corresponding to the highest matching score. If the matching score exceeds the threshold, it indicates that the vehicle is detected and the vehicle appears at the position corresponding to the highest matching score in the scale layer.

The matching score is specifically:

Among them, score is the matching score, (x ₀ ,y ₀ ) is the anchor point, l ₀ is the number of scale layers, is the detection score of the root model, is the response of the i-th component model, v _i is the offset of the i-th component model relative to the anchor point, and b is the offset coefficient.

The model trained by the variable part model algorithm is specifically: by adopting the PCA method to perform feature extraction after the HOG feature dimension reduction processing of the training picture, and then carry out model training through the hidden variable support vector machine classifier, and obtain the variable The model trained by the component model algorithm.

The car judging the distance between the detected vehicle and the own car is specifically: measuring the distance through the laser radar installed on the car to obtain the distance between the detected vehicle and the own car.

The recognition of the lane lines by the car specifically includes: the car scans the collinear points in the image through Hough transform to realize the recognition of the lane lines.

The early warning includes the flashing of the LED early warning panel and the sounding of the buzzer.

The LED early warning panel is provided with LED early warning lights distributed in at least four directions, front, back, left, and right.

Compared with the prior art, the present invention has the following beneficial effects:

(1) Through the variable component model algorithm, the vehicle is recognized on the images captured by the front-view camera and the rear-view camera, and then determined by judging the distance relationship between the vehicle and the vehicle and the position relationship between the vehicle and the lane line Whether the vehicle and the own vehicle are in danger during the overtaking process and whether the own vehicle is in danger during the overtaking process, compared with the existing identification methods through radar or short-distance communication, this method has a faster identification speed and at the same time The accuracy is also relatively high, and the method of identifying through the camera is lower in price and cost than radar and the like.

(2) The vehicle passes through the two steps of the overtaking warning step of the rear vehicle and the overtaking warning step of the front vehicle, and realizes the whole process tracking of the vehicles close to the safe range of the vehicle, that is, the vehicle behind the vehicle passes the front after overtaking The vehicle overtaking warning step is used for monitoring. After overtaking, the vehicle continues to monitor the overtaking vehicle through the rear vehicle overtaking warning step to ensure the integrity of the warning.

(3) When the vehicle is identified and detected by the variable component model algorithm, the scale pyramid is constructed to ensure that the photographed pictures at each scale are recognized, and the picture with the highest detection score is selected as the vehicle recognition position, which ensures the accuracy of the detection. Integrity, thereby improving the accuracy of early warning.

(4) Extract the features of the training pictures and train the model through the variable part model algorithm. The model trained by this method has a higher degree of accuracy, thereby improving the accuracy of the final early warning.

(5) The car recognizes the lane line through Hough transformation, and then can judge whether the car and the warning vehicle are in the same lane line, so that when the two are in the same lane line and the distance is reduced, the driver is warned in time to avoid accidents in time .

(6) The driver is given an early warning by flashing the LED light and the sound of the buzzer at the same time, prompting the driver of the occurrence of dangerous situations to the greatest extent, and improving the safety of the early warning.

(7) The LED early warning panel is equipped with LED early warning lights distributed in at least four directions, front, rear, left, and right, and the position of the detected early warning situation can be indicated by the LED early warning light, so that the driver can know the location of the early warning in time, Improve the timeliness of early warning.

Description of drawings

Fig. 1 is the method flowchart of the overtaking early warning step of vehicle in the rear;

Fig. 2 is the method flowchart of vehicle overtaking early warning step in front;

Fig. 3 is the layout diagram of each module of the automobile realizing the automobile overtaking warning method based on the variable component model algorithm;

FIG. 4 is a schematic diagram of a viewing angle range of a camera;

Fig. 5 is the flow chart of the overall method of overtaking early warning in the embodiment;

Fig. 6 is a model diagram of a variable component model algorithm, where (6a) is the root model diagram of one of the components in the first perspective, (6b) is the component model diagram of one of the components in the first perspective, and (6c) is Model deformation loss in the first viewing angle, (6d) is the root model graph of one of the components in the second viewing angle, (6e) is the part model graph of one of the components in the second viewing angle, (6f) is the root model graph of one of the components in the second viewing angle Model deformation loss;

Fig. 7 is the matching process figure of model;

Figure 8 is a schematic diagram of the LED early warning panel;

Wherein, 111 is a vehicle, 301 is an early warning module, 302 is a front-view camera, 303 is a data processing module, 304 is a rear-view camera, 401-406 are all LED lights.

detailed description

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is carried out on the premise of the technical solution of the present invention, and detailed implementation and specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

What the present embodiment provides is a kind of vehicle overtaking warning method based on variable component model algorithm comprising the following steps:

In the overtaking warning step of the rear vehicle, the car recognizes the rear vehicle through the rear view camera combined with the variable component model algorithm, and by identifying the positional relationship between the rear vehicle, the vehicle and the lane line, judges whether the rear vehicle is overtaking and sends a signal to the driver. Warning, specifically:

A1) The car detects whether other vehicles appear in the image captured by the rear-view camera according to the variable component model algorithm, and if so, proceeds to step A2), otherwise continues to detect:

A11) Build a scale pyramid according to the photographed images of the rear-view camera;

A12) In each scale layer in the scale pyramid, the image is matched with the model trained by the variable component model algorithm by the sliding window detection method, and the matching score is calculated;

A13) Select the scale layer corresponding to the highest matching score, if the matching score exceeds the threshold, it indicates that the vehicle is detected and the vehicle appears at the position corresponding to the highest matching score in the scale layer;

A2) The car judges whether the distance between the detected vehicle and the vehicle decreases, if so, enter step A3), otherwise return to step A1);

A3) The car recognizes the lane line, and judges whether the car and the detected vehicle are located in the same lane line, and if so, sends an early warning to the driver, otherwise proceeds to step A4);

A4) After the detected vehicle disappears from the rear view camera, the car enters the overtaking warning step of the vehicle in front;

In the overtaking warning step of the vehicle in front, the car recognizes the vehicle in front through the front-view camera combined with the variable component model algorithm. By identifying the positional relationship between the vehicle in front, the vehicle and the lane line, it is judged whether the vehicle needs to overtake and whether the overtaking is necessary. Alert the driver in case of danger, specifically:

B1) The car detects whether other vehicles appear in the image taken by the front-view camera according to the variable component model algorithm, if so, enters step B2), otherwise continues to detect;

B2) the car judges whether the distance between the detected vehicle and the vehicle decreases, if so, enter step B3), otherwise return to step B1);

B3) The car recognizes the lane line, and judges whether the car and the detected vehicle are located in the same lane line, and if so, sends an early warning to the driver, otherwise proceeds to step B4);

B4) The car enters the rear vehicle overtaking warning step after the detected vehicle disappears from the front view camera.

The specific process of the above steps is as follows:

As shown in Figure 3, it is the layout diagram of each module of the car that realizes the car overtaking warning method based on the variable component model algorithm. The front view camera 302 is installed in the middle of the top of the front windshield, and is used to collect the road conditions of the 160° angle of view in front of the vehicle 111. information, the rear view camera 304 is installed in the middle of the rear windshield top, and is used to collect the road condition information of the 160° angle of view behind the vehicle 111. The data processing module 303 analyzes and calculates the image information of the front view camera 302 and the rear view camera 304, and recognizes them through calculation The other vehicles and lane lines in the image are tracked, and the vehicles in the field of view are tracked. When the overtaking and lane changing of other adjacent vehicles affect the safe driving of the vehicle 111, the data processing module 303 sends information to the early warning module 301, and the early warning module 301 An alarm is issued, and the corresponding situation is reflected on the LED early warning panel. The early warning module 301 is composed of a buzzer and an LED early warning panel, and is placed on the center console. The viewing angle range of the camera is as shown in Figure 4. In this embodiment, the viewing angle ranges of the front-view camera and the rear-view camera are all 160°, and the collected video can identify the vehicles and vehicles in the adjacent two lanes within the effective range. Lane marking information.

Based on the above components, the vehicle overtaking warning method based on the variable component model algorithm finally proposed is shown in Figure 5, the main steps are: the data processing module 303 detects that a vehicle is coming from the rear through the information of the rear-view camera 304, and The approaching vehicle is tracked, and then it is judged whether the distance between the vehicle and the vehicle is shortened, and if the vehicle is far away from the vehicle, the tracking is abandoned. If the distance between the vehicle and the vehicle is shortened to within a safe distance, the data processing module 303 can judge whether the vehicle is in the same lane line as the vehicle by identifying the lane line and the vehicle. If it is within the lane line, the alarm module will send an alarm to the driver, reminding the driver that a rear-end collision accident may occur. If the vehicle is not within the same lane line as the vehicle, continue to track the vehicle. When the vehicle disappears from the view of the rear-view camera, it means that the vehicle is in the blind spot of the vehicle, and the early warning module sends out an alarm to remind the driver Do not change lanes if other vehicles are located in the blind spot of your own vehicle. The vehicle continues to move forward and appears in the field of view of the front-view camera 302. The digital processing module 303 detects the vehicle through the front-view camera 302 and continues to track the vehicle. If the vehicle is far away from the vehicle in the adjacent lane, Then stop tracking. If the vehicle changes lanes from the adjacent lane to the lane where the vehicle is located, the early warning module will send an alarm to the driver to remind the driver to change lanes of the vehicle ahead and pay attention to slow down and avoid.

The above is the scene where the vehicle is overtaken by other vehicles. The steps in the overtaking other vehicle scene are: the data processing module 303 detects that there is a vehicle in front of the vehicle through the front-view camera 302, and tracks the vehicle. If the distance between the vehicle and the vehicle becomes larger , stop tracking the vehicle. If the distance between the vehicle and the host vehicle is shortened, the data processing module 303 determines whether the vehicle is within the same lane line as the host vehicle. If it is within the same lane line, the early warning module 301 sends an early warning to the driver, reminding the driver that a rear-end collision may occur ahead, and pay attention to slowing down and avoiding. If the vehicle is located in an adjacent lane, then continue to track the vehicle. When disappearing in the front view camera 302 field of view, it means that the vehicle is located in the blind spot of the vehicle, and the early warning module sends an early warning to the driver, reminding the driver that there is a vehicle in the blind spot, and the vehicle cannot change lanes under this request. The vehicle overtakes the vehicle and continues to drive forward. When the vehicle appears in the field of view of the rear view camera 304, the vehicle is tracked. After the vehicle moves away, the vehicle is stopped.

In order to correspond the environmental data collected by the camera with the real objects in the driving environment of the vehicle, and to find the conversion relationship between the point coordinates in the pixel coordinate system of the image generated by the camera and the object point coordinates in the camera environment coordinate system, it is necessary to carry out the camera calibration. This method adopts the joint calibration of the camera and the laser radar, and calibrates the external parameters of the camera by extracting the corresponding feature points of the calibration object on the single-line laser radar and the image, thereby completing the unification of the single-line laser coordinates, camera coordinates, and image pixel coordinates. Enables spatial alignment of lidar and camera.

The joint calibration of the camera and the lidar, the lidar and the camera are rigidly connected to the car, and their respective relative attitudes and displacements are fixed. In the same space, each scanning data point of the lidar has a displacement in the image. corresponding point. Therefore, by establishing a reasonable lidar coordinate system and camera coordinate system, and using the spatial constraint relationship between the lidar scanning point and the camera image, the spatial transformation relationship between the two coordinate systems can be solved, thereby completing the spatial alignment of the lidar and the camera. Realize the correlation of lidar data and visible light image. After the external parameters of the camera are solved through the constraint equation, the relative relationship between the lidar, the camera, the image and the relative environment coordinate system is completely determined, so the lidar scanning point can be projected onto the image pixel coordinates through the camera model head. Its pixel-level data fusion can be completed by the following equation: in, is the internal parameter matrix of the camera. When the camera and the lidar observe point P at the same time, its coordinates in the camera's own environment coordinate system are P _vc (x _vc , y _vc , z _vc ), and the coordinates of the projected point in the visible light image are U=(uv 1) The coordinates of ^T in the radar's own world coordinates are P _lc (x _lc , y _lc , z _lc ). Since the camera and lidar use the same environment coordinate system, there is Where H is the installation height of the lidar. Combining the above two formulas, we can get: in, It can be obtained by the external reference calibration of the lidar and the internal reference calibration of the camera and _Xlc . In summary, by extracting enough image radar corresponding point pairs, the relevant coordinate rotation matrix can be obtained by solving the linear equation and coordinate translation matrix Furthermore, the transformation relationship between the lidar data and its corresponding image pixels can be obtained.

The variable component model algorithm is a target detection algorithm. The variable component model uses PCA (Principal Component Analysis) to perform dimensionality reduction processing on HOG (Histogram of Gradient Oriented) features and then extracts features; in model training, LatentSVM is used The (hidden variable support vector machine) classifier adopts the detection idea of sliding window when detecting the target; for the multi-view problem of the target, it adopts a multi-component strategy to build different models under different perspectives; for the deformation problem of the target, it adopts A component model strategy based on graph structure. Specifically, the variable component model algorithm adopts a star model, which is composed of a root model that generally covers the entire target and a high-resolution component model that covers small components in the target. The root model defines the detection window, and the part model is placed in a feature layer with twice the resolution of the root layer.

The features of the variable part model algorithm adopt the HOG feature, and make some improvements to the HOG feature. The variable part model algorithm cancels the block in the original HOG feature, retains the cell, and directly normalizes an area composed of the current cell and the four surrounding cells when normalizing. When calculating the gradient direction, a strategy combining signed gradient (0-360°) and unsigned gradient (0-180°) is adopted.

The model of the variable component model algorithm, each component is composed of a root filter model and multiple component models. Figure 6(a) and Figure 6(b) are visualizations of the root model and part model of one of the components. As shown in Figure 6(a), the root model is relatively rough, roughly showing the side of a vehicle. As shown in Figure 6(b), the component model is a part inside a rectangular frame, with a total of 6 parts, and the resolution of the component model is twice that of the root model. Figure 6(c) shows the deformation loss of the model. The brighter the area, the greater the cost of deformation loss. The center of the circle is the rational position of the component model. If the detected position of the component model happens to be here, the deformation cost will be 0. The farther the deviation is, the more expensive the deformation will be. For multi-view problems, a multi-component strategy is adopted, and Figure 6(a)(b)(c) and Figure 6(d)(e)(f) are components of two different views respectively.

Algorithm of the variable part model. In terms of algorithm detection, the variable part model adopts a sliding window detection method, and searches at various scales by building a scale pyramid. As shown in Figure 7, the detection process at a certain scale, that is, the matching process of the vehicle model. The model is regarded as a filter operator, and the response score is the degree of similarity between the feature and the model to be matched, and the more similar, the higher the score. As shown in Figure 7, the detection process of the root model is on the left, and the brighter the area, the higher the response score. On the right is the detection process of each component model. First, match the feature image with the model to obtain the filtered image; then, perform the corresponding transformation: take the anchor point (ie, the coordinates of the upper left corner) as the reference position, and integrate the degree of matching between the component model and the feature and the deformation of the component model relative to the ideal position cost (deviation loss), to get the optimal part model position and corresponding score. In the scale l ₀ layer, the detection score with (x ₀ , y ₀ ) as the anchor point is shown in the following formula: in, is the detection score of the root model. Since the same target has multiple components, and the detection scores of different component models need to be aligned, the offset factor b is required. For the response of the i-th component model, since the resolution of the component model is twice that of the root model, the component model needs to be matched at the scale level l ₀ -λ. Therefore, the coordinates of the anchor point also need to be remapped to the scale layer l ₀ -λ, that is, doubled by 2(x ₀ ,y ₀ ), the offset of component model i relative to the anchor point 2(x ₀ ,y ₀ ) The quantity v _i , so in the scale layer l ₀ -λ, the ideal position of component model i is 2(x ₀ ,y ₀ )+v _i .

When hough transform detects a straight line, each pixel coordinate point becomes a set of discrete points after hough transform. Through the discrete polar coordinate formula of a straight line, the geometric equation of the discrete points of the straight line can be expressed as follows: r=x cosθ+y sinθ, where the angle θ refers to the angle between r and the x-axis, and r is the geometric vertical distance to the straight line . We draw each (r, θ) according to the value of the pixel point coordinates (x, y), then convert from the image Cartesian coordinate system to the polar coordinate hough transformation system, this bear point to curve transformation is called the hough transformation of a straight line . The transform quantizes the hough parameter space into a finite number of valued intervals that equally divide or accumulate a grid. At the beginning of the Hough transform algorithm, each pixel coordinate point (x, y) is converted to the curve point (r, θ) and accumulated to the corresponding grid data point. When a peak appears, it means that there is a straight line.

The detection process of the hough transform algorithm is as follows: first, a two-dimensional counter R(r, θ) is established in the parameter space, the range of r is 0 to the length of the diagonal of the image, the range of θ is 0 to 2π, and all values in the array Initialize to 0; then, scan all pixels (x, y) in the image space, the hough transform transforms the image space to the parameter space (r, θ), and the counter R (r, θ) is increased by 1; the third step , set the threshold thr(r, θ), that is, judge how many points in the image are collinear to consider that there is a straight line, and if R(r, θ) is greater than thr(r, θ), the image in the image will be composed.

Fig. 8 is a schematic diagram of the LED early warning panel. There are 6 LED lights on the panel. When other vehicles appear in the blind spot on the left side of the car, the LED light 401 flickers; When other vehicles are in the same lane as the vehicle and are located at the rear of the vehicle, and the distance between the vehicle and the vehicle is too small, the LED light 404 will flash; When the hour is too long, the LED light 403 flickers; when the vehicle in the left lane ahead changes lanes to the same lane as the car, the LED light 402 flickers; when the vehicle in the right lane ahead changes lanes to the same lane as the car, the LED light 405 flickers.

To sum up, this embodiment provides a vehicle overtaking early warning method based on variable component model algorithm. First, the data processing module is trained through the images collected by the front-view camera and the rear-view camera, so that it can use the training to obtain The model can identify the vehicle and the lane line in the actual scene, and track the vehicle, so as to judge other vehicles changing lanes and overtaking around the vehicle, which may cause danger to the vehicle, and send the corresponding scene information to the driver. Early warning, so that the driver can avoid the occurrence of related accidents in advance.

Claims (10)

1. an automobile overtaking warning method based on variable component model algorithm, is characterized in that, described method comprises the following steps: In the overtaking warning step of the rear vehicle, the car recognizes the rear vehicle through the rear view camera combined with the variable component model algorithm, and by identifying the positional relationship between the rear vehicle, the vehicle and the lane line, judges whether the rear vehicle is overtaking and sends a signal to the driver. early warning; In the overtaking warning step of the vehicle in front, the car recognizes the vehicle in front through the front-view camera combined with the variable component model algorithm. By identifying the positional relationship between the vehicle in front, the vehicle and the lane line, it is judged whether the vehicle needs to overtake and whether the overtaking is necessary. Alerts the driver in case of danger. 2. the automobile overtaking early warning method based on variable component model algorithm according to claim 1, is characterized in that, described rear vehicle overtaking early warning step is specifically: A1) The car detects whether other vehicles appear in the image captured by the rear-view camera according to the variable component model algorithm, and if so, enters step A2), otherwise continues to detect; A2) The car judges whether the distance between the detected vehicle and the vehicle decreases, if so, enter step A3), otherwise return to step A1); A3) The car recognizes the lane line, and judges whether the car and the detected vehicle are located in the same lane line, and if so, sends an early warning to the driver, otherwise proceeds to step A4); A4) After the detected vehicle disappears from the rear view camera, the car enters the overtaking warning step of the vehicle in front. 3. the automobile overtaking early warning method based on variable component model algorithm according to claim 1, is characterized in that, described front vehicle overtaking early warning step is specifically: B1) The car detects whether other vehicles appear in the image taken by the front-view camera according to the variable component model algorithm, if so, enters step B2), otherwise continues to detect; B2) the car judges whether the distance between the detected vehicle and the vehicle decreases, if so, enter step B3), otherwise return to step B1); B3) The car recognizes the lane line, and judges whether the car and the detected vehicle are located in the same lane line, if so, an early warning is issued to the driver, otherwise step B4); B4) The car enters the rear vehicle overtaking warning step after the detected vehicle disappears from the front view camera. 4. The automobile overtaking warning method based on variable component model algorithm according to claim 2 or 3, is characterized in that, described variable component model algorithm is specifically: A11) Build a scale pyramid according to the photographed images of the rear-view camera; A12) In each scale layer in the scale pyramid, the image is matched with the model trained by the variable component model algorithm by the sliding window detection method, and the matching score is calculated; A13) Select the scale layer corresponding to the highest matching score. If the matching score exceeds the threshold, it indicates that the vehicle is detected and the vehicle appears at the position corresponding to the highest matching score in the scale layer. 5. the automobile overtaking warning method based on variable component model algorithm according to claim 4, is characterized in that, described matching score is specifically: <mrow> <mi>s</mi> <mi>c</mi> <mi>o</mi> <mi>r</mi> <mi>e</mi> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>,</mo> <msub> <mi>y</mi> <mn>0</mn> </msub> <mo>,</mo> <msub> <mi>l</mi> <mn>0</mn> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>R</mi> <mrow> <mn>0</mn> <mo>,</mo> <msub> <mi>l</mi> <mn>0</mn> </msub> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>,</mo> <msub> <mi>y</mi> <mn>0</mn> </msub> <mo>)</mo> </mrow> <mo>+</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>D</mi> <mrow> <mi>i</mi> <mo>,</mo> <msub> <mi>l</mi> <mn>0</mn> </msub> <mo>-</mo> <mi>&amp;lambda;</mi> </mrow> </msub> <mrow> <mo>(</mo> <mn>2</mn> <mo>(</mo> <mrow> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>,</mo> <msub> <mi>y</mi> <mn>0</mn> </msub> </mrow> <mo>)</mo> <mo>+</mo> <msub> <mi>v</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>+</mo> <mi>b</mi> </mrow> Among them, score is the matching score, (x ₀ ,y ₀ ) is the anchor point, l ₀ is the number of scale layers, is the detection score of the root model, is the response of the i-th component model, v _i is the offset of the i-th component model relative to the anchor point, and b is the offset coefficient. 6. the automobile overtaking warning method based on variable component model algorithm according to claim 4, is characterized in that, described model after the training by variable component model algorithm is specifically: by adopting PCA method to the HOG feature of training picture After dimensionality reduction, feature extraction is carried out, and then model training is carried out through the latent variable support vector machine classifier, and the model trained by the variable component model algorithm is obtained. 7. The vehicle overtaking warning method based on the variable component model algorithm according to claim 2 or 3, wherein the distance between the vehicle and the vehicle detected by the vehicle is specifically: by installing on the vehicle The laser radar is used to measure the distance, and the distance between the detected vehicle and the vehicle is obtained. 8. The vehicle overtaking early warning method based on the variable component model algorithm according to claim 2 or 3, wherein the vehicle recognizes the lane lines as follows: the vehicle scans collinear points in the image through Hough transform , to realize the recognition of lane lines. 9. The vehicle overtaking early warning method based on the variable component model algorithm according to claim 1, wherein the early warning includes the flashing of the LED early warning panel and the sounding of the buzzer to realize the early warning. 10. The vehicle overtaking early warning method based on the variable component model algorithm according to claim 9, wherein the LED early warning panel is provided with LED early warning lights distributed in at least four directions, front, rear, left, and right.