CN111860127A - Vehicle detection method and system, computer readable storage medium - Google Patents

Abstract

The present invention provides a vehicle detection method and system, and a computer-readable storage medium, wherein the vehicle detection method includes: receiving video information; determining vehicle image information corresponding to any frame image in the video information through a vehicle detection model, and The image information determines the position information of the wiper of the vehicle to be detected; based on the position information satisfying a preset condition, it is determined that the vehicle to be detected passes the detection. When detecting the vehicle video uploaded by the driver, the technical solution provided by the present invention determines whether the vehicle video is a real "live" vehicle video according to the motion state of the wiper, which can effectively distinguish cheating methods such as uploading old videos and retaking the screen, and improve the video quality. detection accuracy.

Description

Vehicle detection method and system, computer readable storage medium

technical field

The present invention relates to the technical field of vehicle detection, and in particular, to a vehicle detection method, a vehicle detection system and a computer-readable storage medium.

Background technique

In the related art, for the online car-hailing industry, there is a situation where the vehicle reserved by the user is inconsistent with the vehicle information displayed on the platform. In order to avoid this situation and improve security, the platform needs to verify whether the vehicle driven by the driver is consistent with the registered vehicle information. During the verification process, some drivers cheated by submitting previously recorded videos or re-taking the screen.

Therefore, there is an urgent need for a detection method to ensure that the vehicle information submitted by the driver is authentic and valid.

SUMMARY OF THE INVENTION

The present invention aims to solve at least one of the technical problems existing in the prior art or related technologies.

To this end, a first aspect of the present invention proposes a vehicle detection method.

A second aspect of the present invention provides a vehicle detection system.

A third aspect of the present invention proposes a computer-readable storage medium.

In view of this, a first aspect of the present invention provides a vehicle detection method, including: receiving video information; determining vehicle image information corresponding to any frame image in the video information through a vehicle detection model, and determining to-be-detected according to the vehicle image information Position information of the wiper of the vehicle; based on the position information meeting the preset condition, it is determined that the vehicle to be detected has passed the detection.

In this technical solution, when the vehicle is detected, the video information uploaded by the driver is firstly received. Among them, when the driver is required to shoot the vehicle video, the wiper is turned on at the same time. If the normal operation of the wiper in the video uploaded by the driver is detected, the video uploaded by the driver is considered to be a real-time video of the vehicle. If it is detected that the wiper does not operate normally in the video uploaded by the driver, it indicates that the driver has a high possibility of cheating, and the vehicle in the uploaded video is not a real "live" vehicle, and the vehicle detection fails at this time.

Specifically, the received video information is input into the vehicle detection model, and the video information is processed by the vehicle detection model to obtain continuous frame images, and the frame images of each frame are further processed into corresponding vehicle picture information, and each frame is processed one by one. The frame will determine the position information of the vehicle's wipers in the vehicle image information. When the wiper meets the preset conditions built in the vehicle detection model, it is determined that the wiper operates normally as required, and at this time, it is determined that the vehicle to be detected passes the detection.

When detecting the vehicle video uploaded by the driver, the technical solution provided by the present invention determines whether the vehicle video is a real "live" vehicle video according to the motion state of the wiper, which can effectively distinguish cheating methods such as uploading old videos and retaking the screen, and improve the video quality. detection accuracy.

In addition, the vehicle detection method in the above technical solution provided by the present invention may also have the following additional technical features:

In the above technical solution, the wiper includes a rotating arm and a brush body, and the vehicle to be detected further includes a front windshield; the step of determining the position information of the wiper of the vehicle to be detected according to the vehicle image information specifically includes: determining through a vehicle detection model The first intersection point of the side where the rotating arm and the front windshield intersect, determine the second intersection point between the rotating arm and the brush body, and determine the first end point of the brush body; determine according to the first intersection point, the second intersection point and the first end point location information.

In this technical solution, the wiper of a general vehicle includes a rotating arm and a brush body. After being recognized by the vehicle detection model, the front windshield of the vehicle is recognized as an approximate rectangular area, and the wiper moves within the rectangular area. . Specifically, the point where the wiper intersects with the bottom edge of the corresponding rectangular area of the front windshield is the root node of the wiper, that is, the first intersection, and the wiper approximately rotates around the root node. The position where the rotating arm of the wiper and the brush body intersect is the second intersection point, and the brush body rotates relative to the rotating arm around the second intersection point. At the same time, the brush body itself is processed into the shape of an approximate line segment, and its two ends are marked as the first end point and the second end point, respectively.

Thereby, it can be obtained that the first intersection point and the second intersection point are on the same straight line, that is, they are both located on the rotating arm of the wiper. The second intersection point and the first end point are on the same straight line, that is, the brush body of the wiper. Therefore, the positions of the rotating arm and the brush body of the wiper can be determined through the first intersection point, the second intersection point and the first end point, thereby obtaining the position information of the wiper.

In any of the above technical solutions, the step of determining the position information according to the first intersection, the second intersection and the first endpoint specifically includes: determining the first corner and the second corner corresponding to the front windshield through the vehicle detection model. , the third corner and the fourth corner; take any one of the first corner, the second corner, the third corner and the fourth corner as the origin to establish a Cartesian coordinate system, and take the first intersection as the origin to establish the corner coordinates In the angular coordinate system, determine the data of the included angle between the side where the front windshield intersects the turning arm and the turning arm; in the Cartesian coordinate system, determine the first coordinate corresponding to the first endpoint and the first coordinate corresponding to the second intersection. Two coordinates; the position information is determined according to the angle data, the first coordinates and the second coordinates.

In this technical solution, since the wiper as a whole moves within the approximate rectangle formed by the windshield, the coordinates of the corner points of the four corners of the windshield and the intersection of the wiper and the rectangular area, that is, the first intersection point Coordinates are used as datums to establish a Cartesian coordinate system and an angular coordinate system.

The origin of the Cartesian coordinate system and the pixel length of the unit scale are determined by the four corner points of the windshield, and the origin of the corner coordinate system is the root node of the wiper, that is, the first intersection point. At this time, the first coordinate (x1, y1) and the second coordinate (x2, y2) of the second intersection and the first endpoint in the Cartesian coordinate system are determined by the vehicle identification model, and the wiper arm and the windshield are calculated. The angle alph under the glass, the position of the wiper can be expressed as: (alph, x1, y1, x2, y2), that is, the position information of the wiper.

In any of the above technical solutions, the vehicle detection method further includes: obtaining first position information corresponding to the first frame image in the video information, and obtaining second position information corresponding to other frame images in the video information; determining any second position The first difference between the information and the first position information; based on the existence of at least two first differences that are not equal to each other, and any one of the first differences is greater than a preset threshold, it is determined that the position information satisfies the preset condition.

In this technical solution, firstly obtain the position information of the wiper corresponding to the first frame image in the video information, that is, the first position information. After that, the position information of the wiper corresponding to all frame images except the first frame image in the video information, that is, the second position information, is obtained successively, and the difference between the first position information and the second position information is calculated, which is recorded as first difference.

If the first difference value is a changed value, that is, at least two first difference values are not equal, it means that the position of the wiper is different in at least three frames of images, and it can be determined that the wiper is moving. At the same time, among the plurality of first difference values, if there is at least one first difference value greater than the preset preset threshold value, it means that the movement range of the wiper meets the minimum requirements. The information meets the preset conditions, and the current vehicle can pass the verification.

In any of the above technical solutions, before the step of determining the vehicle image information in any frame image in the video information by the vehicle detection model, the vehicle detection method further includes: acquiring a preset detection model and acquiring preset vehicle image information; A training set is generated according to preset vehicle image information, and a preset detection model is trained through the training set to obtain a vehicle detection model.

In this technical solution, in the process of vehicle detection, the vehicle detection model needs to be used to identify the vehicle image and the position of the wiper, so it is necessary to pre-train a vehicle detection model with an accuracy that meets the requirements. Specifically, an untrained preset detection model is firstly set, specifically, a CNN (Convolutional Neural Networks, convolutional neural network) model can be selected, and a training set is generated through the marked preset vehicle image information, and the preset detection model is trained. . The preset vehicle image information includes images of multi-angle, multi-brand and model vehicles, and in these images, the wipers of the vehicle are in different positions, and the position information of the wipers is marked in advance. Through training and learning, the obtained vehicle detection model can accurately identify the position of the wiper in the vehicle image, and then judge whether the wiper moves as required according to the position information of the wiper, and then complete the detection of the vehicle.

In any of the above technical solutions, the step of training the detection model through the training set specifically includes: inputting the training set into the preset detection model, and obtaining the loss value of the preset detection model, until it is determined that the loss value is less than the preset loss value, Determine the vehicle detection model.

In this technical solution, after the input value of the training set is preset to the detection model, there is a difference between the output value of the detection model and the labeled value of the training set. This difference is the loss value of the current preset detection model, which represents The degree of deviation or loss between the prediction of the wiper position information by the preset detection model and the actual position information. When the loss value is reduced to a certain extent, specifically when the loss value of the detection model is smaller than the preset loss value, it means that the detection accuracy of the preset detection model after training has met the requirements, and the trained preset detection model can be Used as a vehicle detection model.

In any of the above technical solutions, the loss value is the sum of the first loss value and the second loss value, and the preset vehicle image information includes preset wiper position information; wherein the predicted position information output by the preset detection model is the same as the preset value. The difference value of the wiper position information is the first loss value, and the second loss value is calculated through the loss function.

In this technical solution, in order to improve the speed and accuracy of training and make the vehicle detection model easier to optimize, the present invention proposes that the loss value includes a first loss value and a second loss value, wherein the first loss value is the preset model output value and the training value. The difference between the set label values, the loss value of this part is determined by the manually calibrated key point coordinates and the predicted key point coordinates. The second loss value is calculated according to the loss function with self-supervision effect. When the sum of the first loss value and the second loss value is less than the preset loss value, it is determined that the preset detection model after training can be used as the vehicle detection model.

In any of the above technical solutions, the third coordinate corresponding to the second end point of the brush body is determined by a preset detection model; the loss function is specifically:

L _self =‖(y2-y1)(x3-x1)-(y3-y1)(x2-x1)‖;

Among them, L _self is the second loss value, (x1, y1) is the first coordinate, (x2, y2) is the second coordinate, and (x3, y3) is the third coordinate.

In this technical solution, the third coordinate corresponding to the second end point of the brush body is obtained. Since the first end point of the brush body, the second end point of the brush body, and the first intersection point of the brush body and the rotating arm are all located on the brush body, these three points are approximately on the same straight line. The present invention relates to the above-mentioned loss function, which is calculated according to the first coordinates (x1, y1), the second coordinates (x2, y2) and the third coordinates (x3, y3) corresponding to the first end point, the first intersection point and the second end point. Calculate the second loss value. In the specific formula, by taking the norm of the first coordinate, the second coordinate and the third coordinate, it is finally judged whether the above three key points are on the same straight line. If the three key points approach the same straight line, the second loss value Lself approaches zero.

A second aspect of the present invention provides a vehicle detection system, comprising: a memory configured to store a computer program; and a processor configured to execute the computer program to implement the vehicle detection method provided in any of the above technical solutions. Therefore, the vehicle detection system includes all the beneficial effects of the vehicle detection method provided in any of the above technical solutions, and details are not described herein again.

A third aspect of the present invention provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the vehicle detection method provided in any of the above technical solutions. Therefore, the computer-readable storage medium includes all the beneficial effects of the vehicle detection method provided in any of the above technical solutions, and details are not described herein again.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of embodiments taken in conjunction with the accompanying drawings, wherein:

FIG. 1 shows a flowchart of a vehicle detection method according to an embodiment of the present invention;

FIG. 2 shows another flowchart of a vehicle detection method according to an embodiment of the present invention;

Fig. 3 shows another flowchart of a vehicle detection method according to an embodiment of the present invention;

FIG. 4 shows a schematic diagram of wiper position recognition in the vehicle detection method according to the embodiment of the present invention;

Fig. 5 shows another schematic diagram of wiper position recognition in a vehicle detection method according to an embodiment of the present invention;

6 shows a schematic structural diagram of a vehicle detection model in a vehicle detection method according to an embodiment of the present invention;

FIG. 7 shows a flowchart of detecting whether a vehicle is a real vehicle in an embodiment of the present invention;

8 shows a schematic diagram of a frame image of an original video in an embodiment of the present invention;

Fig. 9 shows a schematic diagram of vehicle image information after being recognized by a vehicle detection model in an embodiment of the present invention;

FIG. 10 shows a structural block diagram of a vehicle detection system according to an embodiment of the present invention.

Detailed ways

In order to be able to understand the above objects, features and advantages of the present invention more clearly, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments may be combined with each other in the case of no conflict.

Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific details disclosed below. Example limitations.

The vehicle detection method, vehicle detection system, and computer-readable storage medium according to some embodiments of the present invention are described below with reference to FIGS. 1 to 10 .

Example 1

As shown in FIG. 1, in an embodiment of the first aspect of the present invention, a vehicle detection method is provided, including:

Step S102, receiving video information;

Step S104, determining the vehicle image information corresponding to any frame image in the video information through the vehicle detection model, and determining the position information of the wiper of the vehicle to be detected according to the vehicle image information;

Step S106 , based on the location information satisfying a preset condition, it is determined that the vehicle to be detected has passed the detection.

In this embodiment, when the vehicle is detected, the video information uploaded by the driver is first received. Among them, when the driver is required to shoot the vehicle video, the wiper is turned on at the same time. If the normal operation of the wiper in the video uploaded by the driver is detected, the video uploaded by the driver is considered to be a real-time video of the vehicle. If it is detected that the wiper does not operate normally in the video uploaded by the driver, it indicates that the driver has a high possibility of cheating, and the vehicle in the uploaded video is not a real "live" vehicle, and the vehicle detection fails at this time.

Specifically, the received video information is input into the vehicle detection model, and the video information is processed by the vehicle detection model to obtain continuous frame images, and the frame images of each frame are further processed into corresponding vehicle picture information, and each frame is processed one by one. The frame will determine the position information of the vehicle's wipers in the vehicle image information. When the wiper meets the preset conditions built in the vehicle detection model, it is determined that the wiper operates normally as required, and at this time, it is determined that the vehicle to be detected passes the detection.

When detecting the vehicle video uploaded by the driver, the embodiment of the present invention determines whether the vehicle video is a real "live" vehicle video according to the motion state of the wiper, which can effectively distinguish and upload old videos, remake the screen and other cheating methods, and improve the video quality. detection accuracy.

In an embodiment of the present invention, the wiper includes a rotating arm and a brush body, and the vehicle to be detected further includes a front windshield; as shown in FIG. 2 , the step of determining the position information of the wiper of the vehicle to be detected according to the vehicle image information , including:

Step S202, determining the first intersection point of the side where the rotating arm and the front windshield intersect through the vehicle detection model, determining the second intersection point of the rotating arm and the brush body, and determining the first end point of the brush body;

Step S204, determining the position information according to the first intersection, the second intersection and the first endpoint.

In this embodiment, the wiper of a general vehicle includes a rotating arm and a brush body. After being recognized by the vehicle detection model, the front windshield of the vehicle is recognized as an approximate rectangular area, and the wiper moves within the rectangular area. . Specifically, the point where the wiper intersects with the bottom edge of the corresponding rectangular area of the front windshield is the root node of the wiper, that is, the first intersection, and the wiper approximately rotates around the root node. The position where the rotating arm of the wiper and the brush body intersect is the second intersection point, and the brush body rotates relative to the rotating arm around the second intersection point. At the same time, the brush body itself is processed into the shape of an approximate line segment, and its two ends are marked as the first end point and the second end point, respectively.

Thereby, it can be obtained that the first intersection point and the second intersection point are on the same straight line, that is, they are both located on the rotating arm of the wiper. The second intersection point and the first end point are on the same straight line, that is, the brush body of the wiper. Therefore, the positions of the rotating arm and the brush body of the wiper can be determined through the first intersection point, the second intersection point and the first end point, thereby obtaining the position information of the wiper.

In an embodiment of the present invention, the step of determining the position information according to the first intersection, the second intersection and the first endpoint specifically includes: determining the first corner and the second corner corresponding to the front windshield through a vehicle detection model. point, the third corner point and the fourth corner point; take any one of the first corner point, the second corner point, the third corner point, and the fourth corner point as the origin to establish a Cartesian coordinate system, and use the first intersection as the origin to establish the angle Coordinate system; determine the angle data between the side where the front windshield intersects the rotating arm and the rotating arm in the angular coordinate system; determine the first coordinate corresponding to the first endpoint and the second intersection in the Cartesian coordinate system. The second coordinate; the position information is determined according to the included angle data, the first coordinate and the second coordinate.

In this embodiment, since the wiper as a whole moves within the approximate rectangle formed by the windshield, the coordinates of the corner points of the four corners of the windshield and the intersection of the wiper and the rectangular area, that is, the first intersection point Coordinates are used as datums to establish a Cartesian coordinate system and an angular coordinate system.

The origin of the Cartesian coordinate system and the pixel length of the unit scale are determined by the four corner points of the windshield, and the origin of the corner coordinate system is the root node of the wiper, that is, the first intersection point. At this time, the first coordinate (x1, y1) and the second coordinate (x2, y2) of the second intersection and the first endpoint in the Cartesian coordinate system are determined by the vehicle identification model, and the wiper arm and the windshield are calculated. The angle alph under the glass, the position of the wiper can be expressed as: (alph, x1, y1, x2, y2), that is, the position information of the wiper.

In an embodiment of the present invention, as shown in FIG. 3 , the vehicle detection method further includes:

Step S302, obtaining first position information corresponding to the first frame image in the video information, and obtaining second position information corresponding to other frame images in the video information;

Step S304, determining the first difference between any second position information and the first position information;

Step S306, based on the existence of at least two first differences that are not equal to each other, and any one of the first differences is greater than a preset threshold, determine that the location information satisfies the preset condition.

In this embodiment, the position information of the wiper corresponding to the first frame image in the video information, that is, the first position information, is obtained first. After that, the position information of the wiper corresponding to all frame images except the first frame image in the video information, that is, the second position information, is obtained successively, and the difference between the first position information and the second position information is calculated, which is recorded as first difference.

If the first difference value is a changed value, that is, at least two first difference values are not equal, it means that the position of the wiper is different in at least three frames of images, and it can be determined that the wiper is moving. At the same time, among the plurality of first difference values, if there is at least one first difference value greater than the preset preset threshold value, it means that the movement range of the wiper meets the minimum requirements. The information meets the preset conditions, and the current vehicle can pass the verification.

In an embodiment of the present invention, before the step of determining the vehicle image information in any frame image in the video information by using the vehicle detection model, the vehicle detection method further includes: acquiring a preset detection model and acquiring preset vehicle image information ; Generate a training set according to preset vehicle image information, and train a preset detection model through the training set to obtain a vehicle detection model.

In this embodiment, during the vehicle detection process, the vehicle detection model needs to be used to identify the vehicle image and the position of the wiper, so it is necessary to pre-train a vehicle detection model with an accuracy that meets the requirements. Specifically, an untrained preset detection model is firstly set, specifically, a CNN (Convolutional Neural Networks, convolutional neural network) model can be selected, and a training set is generated through the marked preset vehicle image information, and the preset detection model is trained. . The preset vehicle image information includes images of multi-angle, multi-brand and model vehicles, and in these images, the wipers of the vehicle are in different positions, and the position information of the wipers is marked in advance. Through training and learning, the obtained vehicle detection model can accurately identify the position of the wiper in the vehicle image, and then judge whether the wiper moves as required according to the position information of the wiper, and then complete the detection of the vehicle.

In an embodiment of the present invention, the step of training the detection model through the training set specifically includes: inputting the training set into the preset detection model, and obtaining the loss value of the preset detection model, until it is determined that the loss value is less than the preset loss value , determine the vehicle detection model.

In this embodiment, after the input value of the training set is preset to the detection model, there is a difference between the output value of the detection model and the labeled value of the training set. This difference is the loss value of the current preset detection model, which represents The degree of deviation or loss between the prediction of the wiper position information by the preset detection model and the actual position information. When the loss value is reduced to a certain extent, specifically when the loss value of the detection model is smaller than the preset loss value, it means that the detection accuracy of the preset detection model after training has met the requirements, and the trained preset detection model can be Used as a vehicle detection model.

In an embodiment of the present invention, the loss value is the sum of the first loss value and the second loss value, and the preset vehicle image information includes preset wiper position information; wherein the predicted position information output by the preset detection model is the same as the predicted position information output by the preset detection model. The difference value of the wiper position information is set as the first loss value, and the second loss value is calculated by the loss function.

In this embodiment, in order to improve the speed and accuracy of training and make the vehicle detection model easier to optimize, the present invention proposes that the loss value includes a first loss value and a second loss value, wherein the first loss value is the preset model output value and the training value. The difference between the set label values, the loss value of this part is determined by the manually calibrated key point coordinates and the predicted key point coordinates. The second loss value is calculated according to the loss function with self-supervision effect. When the sum of the first loss value and the second loss value is less than the preset loss value, it is determined that the preset detection model after training can be used as the vehicle detection model.

In an embodiment of the present invention, the third coordinate corresponding to the second end point of the brush body is determined by a preset detection model; the loss function is specifically:

L _self =‖(y2-y1)(x3-x1)-(y3-y1)(x2-x1)‖;

Among them, L _self is the second loss value, (x1, y1) is the first coordinate, (x2, y2) is the second coordinate, and (x3, y3) is the third coordinate.

In this embodiment, the third coordinate corresponding to the second endpoint of the brush body is obtained. Since the first end point of the brush body, the second end point of the brush body, and the first intersection point of the brush body and the rotating arm are all located on the brush body, these three points are approximately on the same straight line. The present invention relates to the above-mentioned loss function, which is calculated according to the first coordinates (x1, y1), the second coordinates (x2, y2) and the third coordinates (x3, y3) corresponding to the first end point, the first intersection point and the second end point. Calculate the second loss value. In the specific formula, by taking the norm of the first coordinate, the second coordinate and the third coordinate, it is finally judged whether the above three key points are on the same straight line. If the three key points approach the same straight line, the second loss value Lself approaches zero.

Embodiment 2

In an embodiment of the present invention, the embodiment of the present invention is described by taking a complete detection process for whether a vehicle is a live vehicle as an example:

First, the concept of live vehicle detection is explained. A similar concept is face liveness detection. There are currently two schemes for face liveness detection, interactive liveness detection and silent liveness detection. The interactive liveness detection of the face completes the authentication of the liveness of the face by detecting whether the face blinks, opens the mouth, shakes the head, etc.; the silent liveness detection of the face relies on a large amount of real face data and non-real face data, using the model Learn the features of non-live faces and the features of live faces to complete the authentication of face live bodies.

The live vehicle detection has a similar idea. The present invention proposes to perform live authentication for the vehicle based on whether the change in the wiper position meets the requirements, so as to verify whether the vehicle is alive.

Specifically, a Cartesian coordinate system and an angular coordinate system are established respectively through the four corner points of the front windshield of the vehicle and the root node of the wiper. Among them, the origin of the Cartesian coordinate system and the pixel length of the unit scale are determined by the four corner points of the car window, and the origin of the corner coordinate system is the root node of the wiper.

Specifically, as shown in FIG. 4 , the wiper position can be represented as (alph, x1, y1, x2, y2).

At the same time, as shown in Figures 5 and 6, the present invention uses the three-level cascaded CNN key point detection model to detect the four corner points of the front window, the root node of the wiper, and three key points of the wiper to determine the position of the wiper.

The first-level CNN model in the three-level CNN model adopts a fully convolutional network, which inputs the vehicle image and outputs the window area, the coordinates of the four window corners and the root node coordinates of the wiper.

The input of the second-level CNN model is the window area output by the first-level CNN model, and the output is the window corner coordinates and the root node of the wiper.

The input of the third-level CNN model is the window corner points output by the second-level CNN model and the aligned window area determined by the root node of the wiper, and the output is the three key points of the wiper.

The key point detection model of the entire three-level cascade CNN detects image information from coarse to fine, and from easy to difficult. The first-level CNN model detects the coarse information of the window area, and the second-level CNN model detects the fine information such as the corners of the window and the root node of the wiper. The third-level CNN model detects the fine information of the three key points of the wiper. The three-level CNN cascade network decouples the detection of fixed key points such as window corners and wiper root nodes from the detection of non-fixed key points such as wiper key points, which makes the model easier to optimize.

Among them, corner regression is also added to the first-level CNN model to assist model training, and to give the model more supervision information to obtain better detection performance.

The invention also designs a loss function with self-supervision, so that the wiper position detection model is easier to optimize. Specifically, no matter how the position of the wiper changes, the positions of the three key points on the wiper are always approximately on the same straight line. The present invention adds this prior information to the loss function of the third-level CNN to provide a stronger supervision signal . It is defined as follows:

L _self =‖(y2-y1)(x3-x1)-(y3-y1)(x2-x1)‖;

Among them, L _self is the second loss value, (x1, y1) is the first coordinate, (x2, y2) is the second coordinate, and (x3, y3) is the third coordinate.

The present invention confirms whether the driver can control the vehicle by verifying whether the driver opens the wiper, and then judges whether the vehicle is a living body. The platform issues an instruction to turn on the wipers, and the driver follows the instruction and shoots a video to upload. The invention detects the wiper position of each frame of pictures in the video, and then calculates the distance of the wiper between the pictures to judge whether the wiper is turned on. The specific process is shown in Figure 7:

Step S702, the platform issues an instruction to turn on the wiper;

Step S704, the vehicle verification personnel turn on the wiper according to the instruction and shoot a video to upload;

Step S706, use the vehicle detection model to detect the vehicle area and generate a new video stream;

Step S708, predicting the wiper position of each frame of picture;

Step S710, calculating the distance of the wiper position of each frame of picture;

Step S712, judge whether the wiper is turned on; if yes, go to step S714, otherwise go to step S716;

Step S714, determine to be a real vehicle;

Step S716, it is determined that the vehicle is not a real vehicle.

In step S706 , the frame image of the original video is shown in FIG. 8 , and the vehicle image information obtained after being recognized by the vehicle detection model is shown in FIG. 9 . The obtained new video stream is input to the key point detection model to detect the four corner points of the front window, the root node of the wiper, and the three key points of the wiper in each frame of the picture, and establish a Cartesian coordinate system and an angular coordinate system to calculate the wiper. Location.

In step S712, if the calculated distance is a series of changing values and the maximum distance value is greater than the set threshold, the vehicle is considered to be a real vehicle (living body), otherwise the vehicle is considered an unreal vehicle (non-living body).

Embodiment 3

As shown in FIG. 10 , in one embodiment of the present invention, a vehicle detection system 1000 is provided, including: a memory 1002 configured to store a computer program; a processor 1004 configured to execute the computer program to achieve the above The vehicle detection method provided in any of the embodiments. Therefore, the vehicle detection system 1000 includes all the beneficial effects of the vehicle detection method provided in any of the above embodiments, and details are not described herein again.

Embodiment 4

In one embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, implements the vehicle detection method provided in any of the above-mentioned embodiments. Therefore, the computer-readable storage medium includes all the beneficial effects of the vehicle detection method provided in any of the above-mentioned embodiments, which will not be repeated here.

In the description of the present invention, the term "plurality" refers to two or more than two, unless otherwise expressly defined, the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation described in the drawings Or the positional relationship is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention; term "Connection", "installation", "fixing", etc. should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or through an intermediate medium. indirectly connected. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

In the description of the present invention, the terms "one embodiment," "some embodiments," "a specific embodiment," and the like, mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in the present invention. at least one embodiment or example of . In the present invention, schematic representations of the above terms do not necessarily refer to the same embodiment or instance. Furthermore, the particular features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. a vehicle detection method, is characterized in that, comprises: receive video information; Determine the vehicle image information corresponding to any frame image in the video information by the vehicle detection model, and determine the position information of the wiper of the vehicle to be detected according to the vehicle image information; Based on the location information satisfying a preset condition, it is determined that the vehicle to be detected has passed the detection. 2 . The vehicle detection method according to claim 1 , wherein the wiper comprises a rotating arm and a brush body, and the vehicle to be detected further comprises a front windshield; 3 . The step of determining the position information of the wiper of the vehicle to be detected according to the vehicle image information specifically includes: The vehicle detection model is used to determine the first intersection point of the side where the rotating arm and the front windshield intersect, determine the second intersection point of the rotating arm and the brush body, and determine the first intersection of the brush body endpoint; The location information is determined from the first intersection, the second intersection, and the first endpoint. 3 . The vehicle detection method according to claim 2 , wherein the step of determining the location information according to the first intersection, the second intersection and the first endpoint specifically comprises: 3 . Determine the first corner point, the second corner point, the third corner point and the fourth corner point corresponding to the front windshield by the vehicle detection model; A Cartesian coordinate system is established with any one of the first corner point, the second corner point, the third corner point, and the fourth corner point as the origin, and corner coordinates are established with the first intersection point as the origin Tie; In the angular coordinate system, determine the angle data between the side where the front windshield intersects the rotating arm and the rotating arm; determining a first coordinate corresponding to the first endpoint and a second coordinate corresponding to the second intersection in the Cartesian coordinate system; The location information is determined according to the included angle data, the first coordinates and the second coordinates. 4. The vehicle detection method according to claim 3, characterized in that, further comprising: obtaining first position information corresponding to the first frame image in the video information, and obtaining second position information corresponding to other frame images in the video information; determining a first difference between any one of the second position information and the first position information; Based on that there are at least two of the first difference values that are not equal to each other, and any one of the first difference values is greater than a preset threshold, it is determined that the location information satisfies the preset condition. 5 . The vehicle detection method according to claim 3 , wherein, before the step of determining the vehicle image information in any frame image in the video information through the vehicle detection model, the vehicle detection method further comprises: 6 . : Obtain the preset detection model and obtain the preset vehicle image information; A training set is generated according to the preset vehicle image information, and the preset detection model is trained through the training set to obtain the vehicle detection model. 6. The vehicle detection method according to claim 5, wherein the step of training the detection model through the training set specifically comprises: The training set is input into the preset detection model, and the loss value of the preset detection model is obtained, until it is determined that the loss value is less than the preset loss value, and the vehicle detection model is determined to be obtained. 7 . The vehicle detection method according to claim 6 , wherein the loss value is the sum of the first loss value and the second loss value, and the preset vehicle image information includes preset wiper position information; 8 . Wherein, the difference between the predicted position information output by the preset detection model and the preset wiper position information is the first loss value, and the second loss value is calculated by using a loss function. 8. The vehicle detection method according to claim 7, further comprising: Determine the third coordinate corresponding to the second end point of the brush body by the preset detection model; The loss function is specifically: L _self =‖(y2-y1)(x3-x1)-(y3-y1)(x2-x1)‖; Wherein, L _self is the second loss value, (x1, y1) is the first coordinate, (x2, y2) is the second coordinate, and (x3, y3) is the third coordinate. 9. A vehicle detection system, comprising: a memory configured to store the computer program; A processor configured to execute the computer program to implement the vehicle detection method as claimed in any one of claims 1 to 8. 10. A computer-readable storage medium on which a computer program is stored, characterized in that, when the computer program is executed by a processor, the vehicle detection method according to any one of claims 1 to 8 is implemented.

Images