Abnormal driving behavior detection method, training method, device and storage medium

Abstract

The embodiments of the present disclosure provide an abnormal driving behavior detection method, training method, device and storage medium. The abnormal driving behavior detection method includes: obtaining driving video data of a target vehicle and road data of a target road; determining driving parameters of the target vehicle based on the driving video data of the target vehicle and the road data of the target road; detecting whether the driver of the target vehicle has abnormal driving behavior based on the driving parameters of the target vehicle and a pre-trained abnormal driving behavior detection model, wherein the abnormal driving behavior detection model is trained by driving parameters of sample vehicles and their corresponding labeled data.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to, but are not limited to, the field of intelligent transportation technology, and in particular, to an abnormal driving behavior detection method, training method, device, and storage medium. Background Art

[0002] With the rapid development of my country's economy and technology, the rapid development of highways, the increasing living standards of people, and the rapid increase in the number of vehicles have brought severe challenges to traffic safety. The hidden dangers of driving safety caused by abnormal driving behaviors of drivers (such as dangerous driving, fatigue driving, etc.) are likely to cause frequent traffic accidents, resulting in huge property losses and casualties. Therefore, detecting abnormal driving behaviors is very important to reduce traffic accidents. Summary of the invention

[0003] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0004] In a first aspect, an embodiment of the present disclosure provides an abnormal driving behavior detection method, comprising: obtaining driving video data of a target vehicle and road data of a target road; determining driving parameters of the target vehicle based on the driving video data of the target vehicle and the road data of the target road; and detecting whether the driver of the target vehicle has abnormal driving behavior based on the driving parameters of the target vehicle and a pre-trained abnormal driving behavior detection model, wherein the abnormal driving behavior detection model is trained by the driving parameters of a sample vehicle and its corresponding labeled data.

[0005] In a second aspect, an embodiment of the present disclosure provides a detection model training method, comprising: obtaining a sample data set; wherein each sample data in the sample data set comprises: driving parameters and labeling data of a sample vehicle; performing training based on the sample data set to obtain an abnormal driving behavior detection model; wherein the abnormal driving behavior detection model is used to detect whether the driver of the target vehicle has abnormal driving behavior based on the driving parameters of the target vehicle.

[0006] In a third aspect, an embodiment of the present disclosure provides a detection device, comprising: a processor and a memory storing a computer program that can be run on the processor, wherein when the processor executes the program, the steps of the abnormal driving behavior detection method as described in the above embodiments are implemented, or the steps of the detection model training method as described in the above embodiments are implemented.

[0007] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, comprising a stored program, wherein when the program is run, the steps of the abnormal driving behavior detection method as described in the above embodiments, or the steps of the detection model training method as described in the above embodiments are implemented.

[0008] The abnormal driving behavior detection method, training method, device and storage medium provided by the embodiments of the present disclosure have the ability to detect abnormal driving behavior because the pre-trained abnormal driving behavior detection model is obtained by training the driving parameters of the sample vehicle and its corresponding labeled data. Then, after determining the driving parameters of the target vehicle through the driving video data of the target vehicle and the road data of the target road, the abnormal driving behavior detection model can be used to detect whether the driver of the target vehicle has abnormal driving behavior. In this way, the accuracy and efficiency of detecting abnormal driving behavior can be improved, and the detection cost can be reduced. Moreover, compared with the technical solution that requires the collection of the driver's physiological parameters or the technical solution that requires the collection of the driver's facial features, since the current vehicles generally integrate driving recorders and positioning systems, it is convenient to obtain the vehicle's driving video data and the road data of the road, which can reduce the detection cost and is not easy to cause user disgust, thereby being conducive to popularization and application in the field of intelligent transportation and traffic safety.

[0009] Other features and advantages of the present disclosure will be described in the following description, and partly become apparent from the description, or be understood by implementing the present disclosure. Other advantages of the present disclosure can be realized and obtained by the schemes described in the description and the drawings.

[0010] Other aspects will be apparent upon reading and understanding the drawings and detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings are used to provide an understanding of the technical solution of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the technical solution of the present disclosure and do not constitute a limitation on the technical solution of the present disclosure. The shapes and sizes of the components in the accompanying drawings do not reflect the actual proportions and are only intended to illustrate the contents of the present disclosure.

[0012] Figure 1 is a flow chart of an abnormal driving behavior detection method in an exemplary embodiment of the present disclosure;

[0013] Figure 2 is a schematic diagram of a training process of a posture estimation model in an exemplary embodiment of the present disclosure;

[0014] Figure 3 is a schematic diagram of a training process of an abnormal driving behavior detection model in an exemplary embodiment of the present disclosure;

[0015] Figure 4 A flowchart of a detection model training method in an exemplary embodiment of the present disclosure;

[0016] Figure 5 It is a schematic diagram of the structure of the detection device in the exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0017] Multiple embodiments are described herein, but the description is exemplary rather than restrictive, and there may be more embodiments and implementations within the scope of the embodiments described herein. Although many possible feature combinations are shown in the drawings and discussed in the specific embodiments, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with any other feature or element in any other embodiment, or may replace any other feature or element in any other embodiment.

[0018] The present disclosure includes and contemplates combinations of features and elements known to those of ordinary skill in the art. The embodiments, features, and elements disclosed in the present disclosure may also be combined with any conventional features or elements to form a unique invention scheme defined by the claims. Any features or elements of any embodiment may also be combined with features or elements from other invention schemes to form another unique invention scheme defined by the claims. Therefore, it should be understood that any feature shown and / or discussed in the present disclosure may be implemented individually or in any appropriate combination. Therefore, except for the limitations made according to the attached claims and their equivalents, the embodiments are not subject to other limitations. In addition, various modifications and changes may be made within the scope of protection of the attached claims.

[0019] When describing representative embodiments, the specification may have presented the method and / or process as a specific sequence of steps. However, to the extent that the method or process does not rely on the specific order of the steps described herein, the method or process should not be limited to the steps of the specific order described. As will be appreciated by those of ordinary skill in the art, other sequences of steps are also possible. Therefore, the specific sequence of the steps set forth in the specification should not be interpreted as a limitation to the claims. In addition, the claims for the method and / or process should not be limited to the steps of performing them in the order written, and those skilled in the art can easily understand that these sequences can be changed and still remain within the spirit and scope of the embodiments of the present application.

[0020] Unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprising" and similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0021] The technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure.

[0022] At present, the technical solutions for detecting abnormal driving behavior can be mainly divided into the following two types: one is mainly to use a camera to collect the driver's facial features (for example, the driver's eye state or the driver's facial expression, etc.) for analysis to determine whether the driver has abnormal driving behavior (for example, fatigue driving). However, products using this technical solution often require the camera to keep shooting the driver's face, which is easy to cause user disgust, and the acceptance level in the market is not high, which is not conducive to popularization and application. Moreover, especially at night when the light in the car is insufficient, it is often very difficult to analyze the driver's facial features, but nighttime is a high-incidence period for abnormal driving behavior (for example, fatigue driving). The other is mainly to collect the driver's physiological parameters for analysis to determine whether the driver has abnormal driving behavior (for example, fatigue driving). However, products using this technical solution often need to collect the driver's physiological parameters in contact, and the detection cost is high, which is not conducive to popularization and application.

[0023] The embodiment of the present disclosure provides a method for detecting abnormal driving behavior. In practical applications, the method for detecting abnormal driving behavior can be applied to a detection device.

[0024] In an exemplary embodiment, the detection device can be set in any scenario where abnormal driving behavior detection services need to be provided, for example, in a vehicle.

[0025] In an exemplary embodiment, the abnormal driving behavior detection method can be applied to various scenarios where the abnormal driving behavior of the driver needs to be detected, such as abnormal driving behavior warning or traffic safety field.

[0026] In an exemplary embodiment, abnormal driving behavior may refer to driving behaviors that are prone to traffic accidents, such as fatigue driving or dangerous driving. For example, factors such as long-term driving, lack of sleep or physiological changes are usually the main causes of fatigue driving. Therefore, when driving while fatigued, the driver may become confused, inattentive, have blurred vision or decreased judgment. In this case, if the driver continues to drive, it is very easy to stop the operation or make correction errors, thus causing traffic accidents. For example, dangerous driving may refer to the vehicle repeatedly making large-angle turns on a long straight road or the vehicle taking too long to react when turning.

[0027] Figure 1 FIG. 1 is a flow chart of an abnormal driving behavior detection method in an exemplary embodiment of the present disclosure, as shown in FIG. Figure 1 As shown, the abnormal driving behavior detection method may include:

[0028] Step 11: Obtain driving video data of the target vehicle and road data of the target road;

[0029] Step 12: Determine driving parameters of the target vehicle based on the driving video data of the target vehicle and the road data of the target road;

[0030] Step 13: Based on the driving parameters of the target vehicle and a pre-trained abnormal driving behavior detection model, detect whether the driver of the target vehicle has abnormal driving behavior; wherein the abnormal driving behavior detection model is trained by the driving parameters of the sample vehicle and its corresponding labeled data.

[0031] In an exemplary embodiment, the driving video data of the target vehicle may refer to video data of the target vehicle's driving process on the target road recorded by a driving recorder in the target vehicle.

[0032] In an exemplary embodiment, the driving video data of the target vehicle may include: at least three frames of images that are continuous in time.

[0033] In an exemplary embodiment, driving video data of the target vehicle may be collected by a driving recorder provided in the target vehicle.

[0034] In an exemplary embodiment, the driving parameters of the target vehicle may include but are not limited to: a first parameter used to characterize the degree of correlation between the driving trajectory of the target vehicle and the linear shape of the target road and a second parameter used to characterize the degree of correlation between the steering angle change data of the target vehicle and the linear angle change data of the target road.

[0035] In an exemplary embodiment, the driving parameters of the sample vehicle may include but are not limited to: a third parameter for characterizing the degree of correlation between the driving trajectory of the sample vehicle and the linear shape of the sample road and a fourth parameter for characterizing the degree of correlation between the steering angle change of the sample vehicle and the linear angle change of the sample road.

[0036] In an exemplary embodiment, the marking data may be artificial marking data for characterizing whether the driver of the sample vehicle has abnormal driving behavior. For example, the marking data may be a Boolean value, including: a preset first value for characterizing that the driver of the sample vehicle has abnormal driving behavior or a preset second value for characterizing that the driver of the sample vehicle does not have abnormal driving behavior. For example, the preset first value may be 1, and the preset second value may be 0.

[0037] As can be seen from the above content, the abnormal driving behavior detection method provided in the exemplary embodiment of the present disclosure has the ability to detect abnormal driving behavior because the pre-trained abnormal driving behavior detection model is obtained by training the driving parameters of the sample vehicle and its corresponding labeled data. Then, after determining the driving parameters of the target vehicle through the driving video data of the target vehicle and the road data of the target road, the abnormal driving behavior detection model can be used to detect whether the driver of the target vehicle has abnormal driving behavior. In this way, the accuracy and efficiency of detecting abnormal driving behavior can be improved, and the detection cost can be reduced. Moreover, compared with the technical solution that requires the collection of the driver's physiological parameters or the technical solution that requires the collection of the driver's facial features, since the current vehicles generally integrate driving recorders and positioning systems, it is convenient to obtain the vehicle's driving video data and the road data of the road, which can reduce the detection cost and is not easy to cause user disgust, thereby being conducive to popularization and application in the field of intelligent transportation and traffic safety.

[0038] In an exemplary embodiment, step 11 may include the following steps 111 to 113:

[0039] Step 111: Acquire driving video data collected by a driving recorder in a target vehicle;

[0040] Step 112: Acquire the location information of the target vehicle collected by the positioning system in the target vehicle;

[0041] Step 113: Based on the location information of the target vehicle, obtain the road data of the target road from the traffic road network.

[0042] In an exemplary embodiment, the positioning system in the target vehicle may refer to a system such as a global positioning system (GPS) that can provide the location information of the vehicle, that is, the location information of the road on which the vehicle is traveling. Here, the embodiments of the present disclosure do not limit this.

[0043] For example, the driving video data of the target vehicle acquired in step 111 may refer to the driving video data of the target vehicle within a predetermined time period collected by the driving recorder, and correspondingly, the position information of the target vehicle acquired in step 112 may refer to multiple position information of the target vehicle within the preset time period. The predetermined time period may be set by the system by default, or may be set by a person skilled in the art. Here, the embodiments of the present disclosure do not limit this.

[0044] In an exemplary embodiment, step 12 may include the following steps 121 to 123:

[0045] Step 121: pre-processing the driving video data of the target vehicle to extract a plurality of target driving image sequences; wherein the target driving image sequence includes: at least three frames of images that are continuous in time;

[0046] Step 122: Processing a plurality of target driving image sequences by means of a trained posture estimation model to obtain motion trajectory data of the target vehicle; wherein the trained posture estimation model is obtained by training an initial self-motion estimation network and an initial depth estimation network by means of the extracted sample driving image sequence;

[0047] Step 123: Determine the driving parameters of the target vehicle based on the motion trajectory data of the target vehicle and the road data of the target road.

[0048] In an exemplary embodiment, step 123 may include the following steps 1231 to 1232:

[0049] Step 1231: Based on the motion trajectory data of the target vehicle and the road data of the target road, a first parameter for characterizing the degree of correlation between the driving trajectory of the target vehicle and the linearity of the target road is calculated; wherein the first parameter may include: a correlation coefficient between the motion trajectory data of the target vehicle and the road data of the target road, an L1 norm of the distance between the motion trajectory data of the target vehicle and the road data of the target road, and an L2 norm of the distance between the motion trajectory data of the target vehicle and the road data of the target road;

[0050] Step 1232: performing differential operations on the motion trajectory data of the target vehicle and the road data of the target road respectively to obtain the steering angle change data of the target vehicle and the linear angle change data of the target road;

[0051] Step 1233: Based on the steering angle change data of the target vehicle and the linear angle change data of the target road, calculate a second parameter for characterizing the degree of correlation between the steering angle change of the target vehicle and the linear angle change of the target road; wherein the second parameter may include: the correlation coefficient between the steering angle change data of the target vehicle and the linear angle change data of the target road, the L1 norm of the angle difference between the steering angle change data of the target vehicle and the linear angle change data of the target road, and the L2 norm of the angle difference between the steering angle change data of the target vehicle and the linear angle change data of the target road.

[0052] In an exemplary embodiment, step 1231 may include: using the motion trajectory data of the target vehicle as a random variable X and the road data of the target road as a random variable Y, and calculating the correlation coefficient between the motion trajectory data of the target vehicle and the road data of the target road through the following formulas (1) to (2); using the distance between the motion trajectory data of the target vehicle and the road data of the target road as a vector x, and calculating the L1 norm of the distance between the motion trajectory data of the target vehicle and the road data of the target road through the following formula (3); using the distance between the motion trajectory data of the target vehicle and the road data of the target road as a vector x, and calculating the L2 norm of the distance between the motion trajectory data of the target vehicle and the road data of the target road through the following formula (4). In this way, a first parameter for characterizing the degree of correlation between the driving trajectory of the target vehicle and the linearity of the target road can be obtained.

[0053] For example, the motion trajectory data of the target vehicle may include: a curve representing the driving trajectory of the target vehicle, and the curve representing the driving trajectory of the target vehicle may include: driving trajectory points in frames. The road data of the target road may include a curve representing the linear shape of the target road. Under normal driving behavior, the driving trajectory points are on the curve representing the linear shape of the target road. At this time, the distance between the motion trajectory data of the target vehicle and the road data of the target road can be 0, and the L1 norm and L2 norm can be 0. When abnormal driving behavior occurs, the driving trajectory points deviate from the target road. At this time, the distance between the motion trajectory data of the target vehicle and the road data of the target road increases, and the L1 norm and L2 norm increase.

[0054] In an exemplary embodiment, step 1233 may include: using the steering angle change data of the target vehicle as the random variable X and the linear angle change data of the target road as the random variable Y, and calculating the correlation coefficient between the steering angle change data of the target vehicle and the linear angle change data of the target road through the following formulas (1) to (2); using the angle difference between the steering angle change data of the target vehicle and the linear angle change data of the target road as the vector x, and calculating the L1 norm of the angle difference between the steering angle change data of the target vehicle and the linear angle change data of the target road through the following formula (3); using the angle difference between the steering angle change data of the target vehicle and the linear angle change data of the target road as the vector x, and calculating the L2 norm of the angle difference between the steering angle change data of the target vehicle and the linear angle change data of the target road through the following formula (4). In this way, a second parameter for characterizing the degree of correlation between the steering angle change of the target vehicle and the linear angle change of the target road can be obtained.

[0055]

[0056] Cov(X,Y)=E(XY)-E(X)E(Y) formula (2);

[0057] ||x|| 1 =|x 1 |+|x 2 |+...+|x n | Formula (3);

[0058]

[0059] Where X and Y are random variables, r(X,Y) is the correlation coefficient between X and Y, Cov(X,Y) is the covariance between X and Y, E(·) is the expectation, Var(·) is the variance, x is a vector, and x 1 to x n is the element in vector x, ||x|| 1 is the L1 norm of x, ||x|| 2 is the L2 norm of x.

[0060] In an exemplary embodiment, step 13 may include: inputting the driving parameters of the target vehicle into a pre-trained abnormal driving behavior detection model to obtain a corresponding output result; and determining whether the driver of the target vehicle has abnormal driving behavior based on the corresponding output result.

[0061] For example, the output result may be a preset first value for characterizing that the driver of the sample vehicle has abnormal driving behavior or a preset second value for characterizing that the driver of the sample vehicle does not have abnormal driving behavior. For example, the preset first value may be 1, and the preset second value may be 0. Here, the embodiments of the present disclosure do not limit this.

[0062] In an exemplary embodiment, before step 11, the method may further include the following steps 14 to 15:

[0063] Step 14: Obtain a sample data set; wherein each sample data in the sample data set includes: driving parameters and marking data of the sample vehicle;

[0064] Step 15: Train based on the sample data set to obtain an abnormal driving behavior detection model. In this way, by training with the driving parameters of the sample vehicle and its corresponding labeled data, a pre-trained abnormal driving behavior detection model can be obtained. The abnormal driving behavior detection model is used to detect whether the driver of the target vehicle has abnormal driving behavior based on the driving parameters of the target vehicle.

[0065] In an exemplary embodiment, before step 11, the method may further include the following step 16: training an initial self-motion estimation network (PoseNet) and an initial depth estimation network (DepthNet) using the extracted sample driving image sequence to obtain a trained posture estimation model.

[0066] In an exemplary embodiment, step 16 may include the following steps 161 to 167:

[0067] Step 161: Acquire driving video data of a sample vehicle;

[0068] Step 162: pre-processing the driving video data of the sample vehicle to extract a plurality of sample driving image sequences; wherein the sample driving image sequence includes: at least three frames of images that are continuous in time;

[0069] Step 163: For each sample driving image sequence, perform the following training operations:

[0070] Step 1631: determining the middle frame image in each sample driving image sequence as the target image, and determining the other frame images except the middle frame image in each sample driving image sequence as the reference image;

[0071] Step 1632: Input the target image and the reference image into an initial self-motion estimation network (PoseNet) to obtain self-motion data of the target image relative to the reference image;

[0072] In an exemplary embodiment, the self-motion data of the target image relative to the reference image may refer to motion estimation data between the target image and the reference image, which may also be referred to as visual odometry (VO).

[0073] Step 1633: Input the target image into the initial depth estimation network (DepthNet) to obtain the depth data of the target image;

[0074] Step 1634: reconstructing the image based on the self-motion data of the target image relative to the reference image and the depth data of the target image to obtain a reconstructed target image;

[0075] Step 1635: Based on the reconstructed target image and the target image, the parameters of the initial ego-motion estimation network and the parameters of the initial depth estimation network are adjusted using a preset loss function;

[0076] Step 164: Perform the above training operation for the next sample driving image sequence until the value of the preset loss function is lower than the preset threshold, and obtain the trained posture estimation model. In this way, the initial self-motion estimation network and the initial depth estimation network are trained by the extracted sample driving image sequence, and the trained posture estimation model can be obtained.

[0077] In an exemplary embodiment, the preset threshold may be set by a person skilled in the art based on experience or experimental results, and the embodiments of the present disclosure do not limit this.

[0078] In an exemplary embodiment, step 166 may include: performing image reconstruction according to the projection relationship shown in the following formula (5) to obtain a reconstructed target image.

[0079] p r ~KE t→r D t (p t )K -1 p t Formula (5);

[0080] Among them, p r is the position of the pixel in the reference image corresponding to the pixel in the target image, K is the intrinsic parameter matrix of the camera, and E t→r is the attitude transfer matrix (i.e., the self-motion data of the target image relative to the reference image), D t (p t ) is the depth data corresponding to the pixel in the target image, K -1 is the inverse matrix of the internal parameter matrix, p t is a pixel in the target image.

[0081] In an exemplary embodiment, the camera intrinsic parameters may be obtained by calibrating the camera. For example, the camera intrinsic parameters may include: the focal length and aperture center of the camera.

[0082] In an exemplary embodiment, the preset loss function may include: any one or more combinations of a photometric consistency loss function, a structural similarity (SSIM) loss function, and an edge-aware smoothness loss function. Here, the embodiments of the present disclosure do not limit this.

[0083] For example, as shown in the following formulas (6) to (9), a weighted sum of a photometric consistency loss function, an SSIM loss function, and an edge-aware smoothness loss function can be used to implement a preset loss function as the objective function of the self-motion estimation network and the depth estimation network to guide the learning process:

[0084]

[0085]

[0086]

[0087]

[0088] Among them, α, β, and γ are weights. is the SSIM loss function, is the photometric loss function, is the edge-aware smoothness loss function, N i is the number of effective pixels when projecting from the reference image to the target image, n is the number of reference images, is the projection of the i-th reference image on the target image plane (i.e., the reconstructed target image), I t is the target image, D t (p) is the depth of point p, is the first-order gradient and e is a constant.

[0089] In an exemplary embodiment, α can be set to 0.85, β can be set to 0.15, and γ can be set to 0.1. Of course, other weight values ​​can also be used, and this is not limited in the embodiments of the present disclosure.

[0090] In an exemplary embodiment, the initial self-motion estimation network may include: multiple convolution blocks, wherein each convolution block may include: a convolution layer and an activation layer, the input of the first convolution block among the multiple convolution blocks is the target image and the reference image, the output of the last convolution block among the multiple convolution blocks is the output of the initial self-motion estimation network, and the output of the i-th convolution block among the multiple convolution blocks is the input of the i+1-th convolution block among the multiple convolution blocks, where i is a positive integer.

[0091] In an exemplary embodiment, the convolution layer can be implemented using the conv2d function. Of course, in addition to the listed conv2d function, the convolution layer can also be implemented using other convolution functions, which is not limited in the embodiments of the present disclosure.

[0092] In an exemplary embodiment, the size of the convolution kernel in the convolution layer may be 3×3 or 5×5, etc. Here, the embodiment of the present disclosure does not limit this.

[0093] In an exemplary embodiment, the initial self-motion estimation network may include: seven convolution blocks, wherein the number of channels of the convolution layers in the seven convolution blocks may be set to 16, 32, 64, 128, 256, 256, and 256, respectively. Of course, it may also be set to other numbers of channels, which is not limited in the embodiments of the present disclosure.

[0094] In an exemplary embodiment, the activation layer of each convolution block can be implemented using a rectified linear unit (ReLU) function. Of course, the activation layer can also be implemented using other activation functions, which is not limited in the embodiments of the present disclosure.

[0095] In an exemplary embodiment, the initial depth estimation network may include: an encoder and a decoder, wherein the encoder may be implemented using a residual network (Residual Network, ResNet), and the decoder may include: a deconvolution layer, the input of the encoder is a reference image, the output of the encoder is the input of the decoder, the output of the encoder is the output of the initial depth estimation network, and skip connections may be added between the encoder and the decoder for transmission.

[0096] In an exemplary embodiment, the structure of the encoder may adopt a typical residual network (ResNet) structure such as ResNet50, ResNet101, etc. Of course, in addition to the residual network (ResNet) listed, the encoder may also adopt other neural network structures. Here, the embodiments of the present disclosure do not limit this.

[0097] In an exemplary embodiment, taking the mini-batch gradient descent (MBGD) as an optimization algorithm when iteratively adjusting the parameters of the initial depth estimation network (DepthNet), the depth network (DepthNet) can learn the target image in a mini-batch manner, that is, after obtaining multiple target images to be trained, the multiple target images can be divided into multiple mini-batch data, and then each mini-batch data is learned, and the parameters of the depth estimation network (DepthNet) are updated in batches, so that a group of data in a batch jointly determines the direction of this gradient. The depth network (DepthNet) can adopt an encoder (encoder) + decoder (decoder) structure, the encoder part can be based on ResNet50 as the basic structure, the decoder part can be composed of a deconvolution layer, and the skip connection (skip connections) is used at the same resolution between the encoder (encoder) and the decoder (decoder) for multi-scale prediction. In this way, global high-level features and local detail information can be retained at the same time.

[0098] In an exemplary embodiment, step 14 may include the following steps 141 to 145:

[0099] Step 141: Acquire driving video data of a sample vehicle and road data of a corresponding sample road;

[0100] Step 142: pre-processing the driving video data of the sample vehicle to extract a plurality of sample driving image sequences; wherein the sample driving image sequence includes: at least three frames of images that are continuous in time;

[0101] Step 143: Processing a plurality of sample driving image sequences by means of a trained posture estimation model to obtain motion trajectory data of the sample vehicle; wherein the trained posture estimation model is obtained by training an initial self-motion estimation network and an initial depth estimation network by means of the extracted sample driving image sequences;

[0102] Step 144: Determine driving parameters of the sample vehicle based on the motion trajectory data of the sample vehicle and the road data of the sample road;

[0103] Step 145: Based on the driving parameters of the sample vehicle, obtain corresponding marking data.

[0104] In an exemplary embodiment, there may be multiple sample vehicles and multiple sample roads, which are not limited in the present disclosure.

[0105] In an exemplary embodiment, the sample road may refer to a road itself, or may refer to a section of a road (for example, a road section that meets a preset length and a preset linear shape), etc. Here, the embodiment of the present disclosure does not limit this.

[0106] In an exemplary embodiment, the driving video data of the sample vehicle may refer to video data of the sample vehicle driving process recorded by a driving recorder in the sample vehicle when the sample vehicle is driving on a sample road.

[0107] In an exemplary embodiment, the marking data may be artificial marking data for characterizing whether the driver of the sample vehicle has abnormal driving behavior. For example, the marking data may be a Boolean value, including: a preset first value for characterizing that the driver of the sample vehicle has abnormal driving behavior or a preset second value for characterizing that the driver of the sample vehicle does not have abnormal driving behavior. For example, the preset first value may be 1, and the preset second value may be 0.

[0108] In an exemplary embodiment, the driving parameters of the sample vehicle may include but are not limited to: a third parameter for characterizing the degree of correlation between the driving trajectory of the sample vehicle and the linear shape of the sample road and a fourth parameter for characterizing the degree of correlation between the steering angle change of the sample vehicle and the linear angle change of the sample road.

[0109] In an exemplary embodiment, step 141 may include the following steps 1411 to 1413:

[0110] Step 1411: Obtain driving video data collected by a driving recorder in a sample vehicle;

[0111] Step 1412: Obtaining location information of the sample vehicle collected by a positioning system (eg, GPS) in the sample vehicle;

[0112] Step 1413: Based on the location information of the sample vehicle, obtain road data of the sample road from the traffic road network.

[0113] In an exemplary embodiment, after step 142, the method may further include: preprocessing the sample driving image sequence. For example, the preprocessing may include: any one or more of denoising, averaging, normalization, and image standardization. For example, the denoising may be any one or more of bilateral filtering, Gaussian filtering, binning denoising, KD-Tree (K-Dimensional Tree) denoising, straight-through filtering, random sampling consistency filtering, and denoising based on statistical methods. Here, the embodiments of the present disclosure are not limited to this.

[0114] In an exemplary embodiment, step 144 may include the following steps 1441 to 1443:

[0115] Step 1441: Based on the motion trajectory data of the sample vehicle and the road data of the sample road, a third parameter for characterizing the degree of correlation between the driving trajectory of the sample vehicle and the linearity of the sample road is calculated; wherein the third parameter may include: a correlation coefficient between the motion trajectory data of the sample vehicle and the road data of the sample road, an L1 norm of the distance between the motion trajectory data of the sample vehicle and the road data of the sample road, and an L2 norm of the distance between the motion trajectory data of the sample vehicle and the road data of the sample road;

[0116] Step 1442: performing differential operations on the motion trajectory data of the sample vehicle and the road data of the sample road respectively to obtain the steering angle change data of the sample vehicle and the linear angle change data of the sample road;

[0117] Step 1443: Based on the steering angle change data of the sample vehicle and the linear angle change data of the sample road, calculate and obtain a fourth parameter for characterizing the degree of correlation between the steering angle change of the sample vehicle and the linear angle change of the sample road; wherein the fourth parameter may include: the correlation coefficient between the steering angle change data of the sample vehicle and the linear angle change data of the sample road, the L1 norm of the angle difference between the steering angle change data of the sample vehicle and the linear angle change data of the sample road, and the L2 norm of the angle difference between the steering angle change data of the sample vehicle and the linear angle change data of the sample road.

[0118] In an exemplary embodiment, step 1441 may include: based on the motion trajectory data of the sample vehicle and the road data of the sample road, calculating the third parameter through the above formulas (1) to (4). The third parameter can be calculated through the above formulas (1) to (4). The process of calculating the third parameter can be understood by referring to the description of the process of calculating the first parameter above, and will not be described in detail here.

[0119] Step 1443 may include: calculating a fourth parameter based on the steering angle change data of the sample vehicle and the linear angle change data of the sample road by using the above formulas (1) to (4). Regarding the process of calculating the fourth parameter, those skilled in the art may refer to the description of the process of calculating the second parameter above for understanding, and will not be described in detail here.

[0120] Figure 2 is a schematic diagram of a training process of a posture estimation model in an exemplary embodiment of the present disclosure, Figure 3 FIG. 1 is a schematic diagram of the training process of the abnormal driving behavior detection model in the exemplary embodiment of the present disclosure. Figure 2 and Figure 3 In the example, the posture estimation model includes a self-motion estimation network and a depth estimation network, the self-motion estimation network includes seven convolution blocks, and the depth estimation network includes an encoder and a decoder. Then, the process of training the abnormal driving behavior detection model may include:

[0121] The first step is to obtain a large amount of driving video data of sample vehicles and road data of corresponding sample roads.

[0122] For example, multiple roads can be pre-selected from the traffic road network, and each selected road can be divided into multiple road segments according to a preset strategy (e.g., preset length), so that multiple sample roads can be pre-set. When different sample vehicles are driving on multiple sample roads, for each sample vehicle, the driving video data of the sample vehicle is collected through the driving recorder in the sample vehicle, and the location information of the sample vehicle is collected through the positioning system (e.g., GPS) in the sample vehicle, so as to obtain the road data of the sample road from the traffic road network according to the location information of the sample vehicle. In this way, multiple driving video data of multiple sample vehicles and the corresponding road data of multiple sample roads are obtained.

[0123] Next, based on the multiple driving video data of the multiple sample vehicles, the motion trajectory data of the multiple sample vehicles can be obtained through the following second to fifth steps.

[0124] The second step is to pre-process the driving video data of each sample vehicle and extract multiple sample driving image sequences, wherein the sample driving image sequence may include: at least three frames of images that are continuous in time. In this way, a large number of sample driving image sequences can be obtained.

[0125] In the third step, according to the data volume of the obtained sample driving image sequence, the obtained sample driving image sequence can be divided into three parts, including: training set, verification set and test set, wherein the training set can be used for training in the training phase, and the verification set and the test set can be used to verify the effect of the trained model (if the effect is not good, it can be returned for retraining). In this way, the detection effect of the abnormal driving behavior detection model can be better.

[0126] For example, when the total data volume of the obtained sample driving image sequence is 0 to 10,000, the training set, the validation set, and the test set may be divided according to 7:2:1. Alternatively, when the total data volume of the obtained sample driving image sequence is 10,000 to 1,000,000, the training set, the validation set, and the test set may be divided according to 8:1:1. This is not limited in the embodiments of the present disclosure.

[0127] For example, the KITTI dataset can be used, in which the total data amount of the sample driving image sequences included in the training set can be set to 40109, the total data amount of the sample driving image sequences included in the validation set can be set to 4431, and the total data amount of the sample driving image sequences included in the test set can be set to 697.

[0128] The fourth step is Figure 3 As shown, for each sample driving image sequence, the following training operations are performed: the middle frame image in each sample driving image sequence is determined as the target image; the other frame images except the middle frame image in each sample driving image sequence are determined as reference images; the target image and the reference image are input into the initial self-motion estimation network (PoseNet) to obtain the self-motion data of the target image relative to the reference image; the target image is input into the initial depth estimation network (DepthNet) to obtain the depth data of the target image; image reconstruction is performed based on the self-motion data of the target image relative to the reference image and the depth data of the target image to obtain a reconstructed target image; based on the reconstructed target image and the target image, the preset loss function is used to adjust the parameters of the initial self-motion estimation network and the parameters of the initial depth estimation network; the above training operations are performed for the next sample driving image sequence until the value of the preset loss function is lower than the preset threshold, and the trained posture estimation model is obtained.

[0129] In this way, the initial self-motion estimation network and the initial depth estimation network are trained by extracting the driving image sequence, and a trained posture estimation model can be obtained.

[0130] After the training is completed, the validation set and the test set can be used to test and verify the trained posture estimation model to determine whether the model effect of the trained posture estimation model meets the preset requirements.

[0131] Step 5: Figure 3 As shown, for each sample driving image sequence, the following operations are performed: the middle frame image in each sample driving image sequence is determined as the target image; the other frame images except the middle frame image in each sample driving image sequence are determined as reference images; the target image and the reference image are input into the trained posture estimation model to obtain the motion trajectory data of the sample vehicle.

[0132] The sixth step is to calculate a third parameter for characterizing the degree of correlation between the driving trajectory of the sample vehicle and the linearity of the sample road based on the motion trajectory data of the sample vehicle and the road data of the sample road; wherein the third parameter may include: a correlation coefficient between the motion trajectory data of the sample vehicle and the road data of the sample road, an L1 norm of the distance between the motion trajectory data of the sample vehicle and the road data of the sample road, and an L2 norm of the distance between the motion trajectory data of the sample vehicle and the road data of the sample road.

[0133] The seventh step is to perform differential operations on the motion trajectory data of the sample vehicle and the road data of the sample road respectively to obtain the steering angle change data of the sample vehicle and the linear angle change data of the sample road; based on the steering angle change data of the sample vehicle and the linear angle change data of the sample road, calculate and obtain a fourth parameter for characterizing the degree of correlation between the steering angle change of the sample vehicle and the linear angle change of the sample road; wherein the fourth parameter may include: the correlation coefficient between the steering angle change data of the sample vehicle and the linear angle change data of the sample road, the L1 norm of the angle difference between the steering angle change data of the sample vehicle and the linear angle change data of the sample road, and the L2 norm of the angle difference between the steering angle change data of the sample vehicle and the linear angle change data of the sample road.

[0134] For example, the third parameter and the fourth parameter are calculated by the above formula (1) to formula (4). Regarding the process of calculating the third parameter and the fourth parameter, those skilled in the art can refer to the above description of the process of calculating the first parameter and the second parameter for understanding, and the embodiments of the present disclosure will not be described in detail here.

[0135] So, like Figure 3 As shown, a large number of driving parameters of sample vehicles can be obtained through the motion trajectory data of sample vehicles and the road data of sample roads.

[0136] The eighth step is to manually mark the driving parameters of the sample vehicles, so that the corresponding marked data can be obtained.

[0137] Step 9: Figure 3As shown, the driving parameters of the sample vehicle and its corresponding labeled data are encapsulated into new data and trained to obtain an abnormal driving behavior detection model. In this way, an abnormal driving behavior detection model can be obtained. The abnormal driving behavior detection model is used to detect whether the driver of the target vehicle has abnormal driving behavior based on the driving parameters of the target vehicle.

[0138] In an exemplary embodiment, machine learning (ML) techniques such as deep neural networks (DNN) or convolutional neural networks (CNN) can be used to train new data obtained by encapsulating the driving parameters of sample vehicles and their corresponding labeled data to obtain an abnormal driving behavior detection model.

[0139] At this point, the process of training the abnormal driving behavior detection model is completed.

[0140] The disclosed embodiment also provides a detection model training method. In practical applications, the detection model training method can be used to train the abnormal driving behavior detection model in one or more of the above embodiments.

[0141] Figure 4 FIG. 1 is a flow chart of a detection model training method in an exemplary embodiment of the present disclosure, such as Figure 4 As shown, the detection model training method may include:

[0142] Step 41: obtaining a sample data set; wherein each sample data in the sample data set includes: driving parameters and marking data of the sample vehicle;

[0143] Step 42: Perform training based on the sample data set to obtain an abnormal driving behavior detection model; wherein the abnormal driving behavior detection model is used to detect whether the driver of the target vehicle has abnormal driving behavior based on the driving parameters of the target vehicle.

[0144] The description of the above detection model training method embodiment is similar to the description of the above abnormal driving behavior detection method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the detection model training method embodiment of the present disclosure, technicians in this field should refer to the description in the abnormal driving behavior detection method embodiment of the present disclosure for understanding, and no further description is given here.

[0145] The embodiments of the present disclosure also provide a detection device. The detection device may include: a processor and a memory storing a computer program that can be run on the processor, wherein when the processor executes the computer program, the steps of the abnormal driving behavior detection method in one or more of the above exemplary embodiments are implemented, or the steps of the detection model training method in one or more of the above exemplary embodiments are implemented.

[0146] In an exemplary embodiment, Figure 5 As shown, the detection device 50 may include: at least one processor 51; and at least one memory 52 and a bus 53 connected to the processor 51; wherein the processor 51 and the memory 52 communicate with each other via the bus 53; the processor 51 is used to call the program instructions in the memory 52 to execute the steps of the abnormal driving behavior detection method in one or more of the above-mentioned exemplary embodiments, or the steps of the detection model training method in one or more of the above-mentioned exemplary embodiments.

[0147] In an exemplary embodiment, the processor may be a central processing unit (CPU), other general-purpose processors, a digital signal processor (DSP), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, application-specific integrated circuits, etc. A general-purpose processor may be a microprocessor (MPU) or the processor may be any conventional processor, etc. Here, the embodiments of the present disclosure do not limit this.

[0148] In an exemplary embodiment, the memory may include a non-permanent memory in a computer-readable storage medium, a random access memory (RAM) and / or a non-volatile memory, such as a read-only memory (ROM) or a flash memory (Flash RAM), and the memory includes at least one storage chip. Here, the embodiments of the present disclosure are not limited to this.

[0149] In an exemplary embodiment, the bus may include a power bus, a control bus, a status signal bus, etc. in addition to the data bus. Figure 5 In the figure, various buses are all labeled as bus 53. Here, the embodiment of the present disclosure does not limit this.

[0150] In an exemplary embodiment, the vehicle where the detection device is located may also include, but is not limited to, the following components: a driving recorder or a positioning system, etc. For example, the driving recorder may be configured to collect driving video data of the vehicle. For example, the driving recorder may include at least one camera, etc. For example, the positioning system may be configured to collect location information of the vehicle. For example, the positioning system may be a GPS.

[0151] In the implementation process, the processing performed by the detection device can be completed by the hardware integrated logic circuit in the processor or the instructions in the form of software. That is, the method steps of the embodiment of the present disclosure can be embodied as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0152] The present disclosure also provides a computer-readable storage medium, including a stored program, wherein when the program is run, the steps of the abnormal driving behavior detection method in one or more of the above-mentioned exemplary embodiments or the steps of the detection model training method in one or more of the above-mentioned exemplary embodiments are implemented.

[0153] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, ROM / RAM, a magnetic disk or an optical disk, etc. Here, the embodiment of the present disclosure does not limit this.

[0154] The description of the above detection device or computer-readable storage medium embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the detection device or computer-readable storage medium embodiment of the present disclosure, those skilled in the art refer to the description of the method embodiment of the present disclosure for understanding. No further details are given here.

[0155] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0156] Although the embodiments disclosed in the present disclosure are as above, the above contents are only embodiments adopted for facilitating the understanding of the present disclosure and are not intended to limit the present disclosure. Any technician in the field to which the present disclosure belongs can make any modifications and changes in the form and details of implementation without departing from the spirit and scope disclosed in the present disclosure, but the scope of patent protection of the present disclosure shall still be subject to the scope defined in the attached claims.

Claims

1. A method for detecting abnormal driving behavior, characterized in that: include: Obtain driving video data of a target vehicle and road data of a target road; Determining driving parameters of the target vehicle based on the driving video data of the target vehicle and the road data of the target road; Based on the driving parameters of the target vehicle and a pre-trained abnormal driving behavior detection model, detecting whether the driver of the target vehicle has abnormal driving behavior, wherein the abnormal driving behavior detection model is trained by the driving parameters of the sample vehicle and its corresponding labeled data; The driving parameters of the target vehicle include: a first parameter for characterizing the correlation between the driving trajectory of the target vehicle and the linear shape of the target road, and a second parameter for characterizing the correlation between the steering angle change of the target vehicle and the linear angle change of the target road. The driving parameters of the sample vehicle include: a third parameter for characterizing the correlation between the driving trajectory of the sample vehicle and the linear shape of the sample road, and a fourth parameter for characterizing the correlation between the steering angle change of the sample vehicle and the linear angle change of the sample road.

2. The method according to claim 1, characterized in that The step of obtaining driving video data of a target vehicle and road data of a target road includes: Acquiring the driving video data collected by the driving recorder in the target vehicle; Acquiring location information of the target vehicle collected by a positioning system in the target vehicle; Based on the location information of the target vehicle, road data of the target road is acquired from a traffic road network.

3. The method according to claim 1, characterized in that The determining the driving parameters of the target vehicle based on the driving video data of the target vehicle and the road data of the target road comprises: Preprocessing the driving video data of the target vehicle to extract multiple target driving image sequences; wherein the target driving image sequence includes: at least three frames of images that are continuous in time; The plurality of target driving image sequences are processed by the trained posture estimation model to obtain the motion trajectory data of the target vehicle; wherein the trained posture estimation model is obtained by training an initial self-motion estimation network and an initial depth estimation network by using the extracted sample driving image sequence; Based on the motion trajectory data of the target vehicle and the road data of the target road, the driving parameters of the target vehicle are determined.

4. The method according to claim 3, characterized in that The determining of the driving parameters of the target vehicle based on the motion trajectory data of the target vehicle and the road data of the target road includes: The first parameter is calculated based on the motion trajectory data of the target vehicle and the road data of the target road; wherein the first parameter includes: a correlation coefficient between the motion trajectory data of the target vehicle and the road data of the target road, an L1 norm of a distance between the motion trajectory data of the target vehicle and the road data of the target road, and an L2 norm of a distance between the motion trajectory data of the target vehicle and the road data of the target road; Performing differential operations on the motion trajectory data of the target vehicle and the road data of the target road respectively to obtain the steering angle change data of the target vehicle and the linear angle change data of the target road; The second parameter is calculated based on the steering angle change data of the target vehicle and the linear angle change data of the target road; wherein the second parameter includes: the correlation coefficient between the steering angle change data of the target vehicle and the linear angle change data of the target road, the L1 norm of the angle difference between the steering angle change data of the target vehicle and the linear angle change data of the target road, and the L2 norm of the angle difference between the steering angle change data of the target vehicle and the linear angle change data of the target road.

5. The method according to claim 1, characterized in that The abnormal driving behavior detection model is trained by the driving parameters of the sample vehicles and their corresponding labeled data, including: Obtaining a sample data set; wherein each sample data in the sample data set includes: driving parameters and marking data of a sample vehicle; The abnormal driving behavior detection model is obtained by training based on the sample data set.

6. The method according to claim 5, characterized in that The obtaining of the sample data set comprises: Acquire driving video data of a sample vehicle and road data of a corresponding sample road; Preprocessing the driving video data of the sample vehicle to extract a plurality of sample driving image sequences; wherein the sample driving image sequences include: at least three frames of images that are continuous in time; The plurality of sample driving image sequences are processed by the trained posture estimation model to obtain the motion trajectory data of the sample vehicle; wherein the trained posture estimation model is obtained by training an initial self-motion estimation network and an initial depth estimation network by using the extracted sample driving image sequence; Determining driving parameters of the sample vehicle based on the motion trajectory data of the sample vehicle and the road data of the sample road; Based on the driving parameters of the sample vehicle, corresponding marking data is obtained.

7. The method according to claim 3 or 6, characterized in that: The trained posture estimation model is obtained by training an initial self-motion estimation network and an initial depth estimation network through the extracted sample driving image sequence, including: Obtain driving video data of sample vehicles; Preprocessing the driving video data of the sample vehicle to extract a plurality of sample driving image sequences; wherein the sample driving image sequences include: at least three frames of images that are continuous in time; For each sample driving image sequence, perform the following training operations: Determine the middle frame image in each sample driving image sequence as the target image, and determine the other frame images in each sample driving image sequence except the middle frame image as the reference image; Inputting the target image and the reference image into an initial self-motion estimation network to obtain self-motion data of the target image relative to the reference image; Inputting the target image into an initial depth estimation network to obtain depth data of the target image; Perform image reconstruction based on the self-motion data and the depth data to obtain a reconstructed target image; Based on the reconstructed target image and the target image, adjusting parameters of the initial ego-motion estimation network and parameters of the initial depth estimation network using a preset loss function; The above training operation is performed for the next sample driving image sequence until the value of the preset loss function is lower than the preset threshold, thereby obtaining the trained posture estimation model.

8. The method according to claim 7, characterized in that The initial self-motion estimation network includes: multiple convolution blocks, wherein each convolution block includes: a convolution layer and an activation layer, the input of the first convolution block among the multiple convolution blocks is the target image and the reference image, the output of the last convolution block among the multiple convolution blocks is the output of the initial self-motion estimation network, the output of the i-th convolution block among the multiple convolution blocks is the input of the i+1-th convolution block among the multiple convolution blocks, and i is a positive integer.

9. The method according to claim 8, characterized in that The multiple convolution blocks include: seven convolution blocks, wherein the number of channels of the convolution layers in the seven convolution blocks are 16, 32, 64, 128, 256, 256 and 256 respectively, and the activation layer of each convolution block is implemented using a rectified linear unit function.

10. The method according to claim 7, characterized in that The initial depth estimation network includes: an encoder and a decoder, wherein the encoder is implemented using a residual network, and the decoder includes: a deconvolution layer, the input of the encoder is the reference image, the output of the encoder is the input of the decoder, the output of the encoder is the output of the initial depth estimation network, and a jump connection is added between the encoder and the decoder for transmission.

11. The method according to claim 7, characterized in that The preset loss function includes: any one or more combinations of a photometric consistency loss function, a structural similarity loss function, and an edge-aware smoothness loss function.

12. A detection model training method, comprising: Obtaining a sample data set; wherein each sample data in the sample data set includes: driving parameters and marking data of a sample vehicle; Training is performed based on the sample data set to obtain an abnormal driving behavior detection model; wherein the abnormal driving behavior detection model is used to detect whether the driver of the target vehicle has abnormal driving behavior based on the driving parameters of the target vehicle; The driving parameters of the target vehicle include: a first parameter for characterizing the correlation between the driving trajectory of the target vehicle and the linear shape of the target road, and a second parameter for characterizing the correlation between the steering angle change of the target vehicle and the linear angle change of the target road. The driving parameters of the sample vehicle include: a third parameter for characterizing the correlation between the driving trajectory of the sample vehicle and the linear shape of the sample road, and a fourth parameter for characterizing the correlation between the steering angle change of the sample vehicle and the linear angle change of the sample road.

13. A detection device, comprising: A processor and a memory storing a computer program that can be run on the processor, wherein when the processor executes the program, the steps of the abnormal driving behavior detection method as described in any one of claims 1 to 11 are implemented, or the steps of the detection model training method as described in claim 12 are implemented.

14. A computer-readable storage medium comprising a stored program, wherein: When the program is running, the steps of the abnormal driving behavior detection method as described in any one of claims 1 to 11, or the steps of the detection model training method as described in claim 12 are implemented.

Images