Intelligent identification and classification method and system for traffic dangerous driving scene

Abstract

The invention provides an intelligent identification and classification method and system for a traffic dangerous driving scene. The intelligent identification and classification method for the traffic dangerous driving scene comprises the following steps: extracting each accident characteristic value corresponding to each preset accident characteristic in a traffic accident video; converting each accident characteristic value into a numerical value form; carrying out dimension reduction by utilizing classification principal component analysis; using an elbow method and a K-means clustering method to obtain an optimal clustering number and a clustering result, and dividing the traffic accident video into a plurality of different categories according to the clustering result; and determining the main features of each category and the dangerous driving category corresponding to each category by using statistical analysis. According to the method, different types of traffic dangerous driving scenes can be intelligently identified, and driver risk perception training and scene identification and decision making of an auxiliary automatic driving system can be better carried out.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation systems, and in particular to a method and system for identifying and classifying dangerous traffic driving scenarios. Background Art

[0002] With the rapid development of autonomous driving technology, traffic safety has become one of the key issues. Existing autonomous driving systems use sensors such as vision, radar and lidar to perceive the vehicle's surrounding environment, which has improved driving safety to a certain extent. However, accurate identification and classification of traffic hazard scenes remains a challenge, especially in complex and dynamic traffic environments. Existing technologies have shown certain limitations in dealing with diverse dangerous driving behaviors.

[0003] Existing autonomous driving systems achieve basic safety functions through simple obstacle detection, vehicle tracking or pedestrian detection, but when encountering complex traffic situations, especially dangerous traffic scenarios involving pedestrians crossing the street and vehicles changing lanes illegally, these systems are usually not capable of identifying them. Existing methods often rely on single sensor data or overly simple rules, lacking comprehensive analysis of complex interactions of multiple factors, resulting in an inability to effectively respond to potential risks in scenarios such as highways, urban roads or complex intersections. Summary of the invention

[0004] In order to solve the deficiencies of the prior art, the present invention proposes a method and system for intelligently identifying and classifying traffic hazardous driving scenarios.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for intelligently identifying and classifying dangerous traffic driving scenes, characterized in that: for at least two traffic accident videos, the classification of the traffic accident videos is implemented according to steps S1 to S5:

[0007] Step S1, obtaining a preset number of traffic accident videos, extracting accident feature values ​​corresponding to preset accident features in each traffic accident video using deep learning and computer vision technology, and then executing step S2;

[0008] Step S2, for each accident feature value corresponding to each traffic accident video, in combination with a preset conversion rule, each accident feature value is converted into a numerical form, that is, an accident numerical feature, and the accident numerical feature is standardized, and then step S3 is executed;

[0009] Step S3, based on the accident numerical features corresponding to each traffic accident video, use classification principal component analysis to perform dimensionality reduction to obtain the principal component score, and then execute step S4;

[0010] Step S4, based on the principal component scores, using the elbow method and K-means clustering method, obtain the optimal number of clusters and clustering results, and divide all traffic accident videos into multiple different categories according to the clustering results, and then execute step S5;

[0011] Step S5, based on the traffic accident videos included in each category and in combination with the accident feature values ​​corresponding to each traffic accident video in each category, statistical analysis is used to determine the main features of each category, and then determine the dangerous driving category corresponding to each category.

[0012] Furthermore, the step S1 acquires the traffic accident video from an image capture device carried by a party involved in the accident or a party not involved in the accident.

[0013] Furthermore, the step S3 uses the following steps to perform dimensionality reduction using classification principal component analysis to obtain principal component scores:

[0014] Step S31, constructing an original data matrix based on the accident numerical features corresponding to each traffic accident video, and calculating the covariance matrix of the original data matrix, and then executing step S32;

[0015] Step S32, calculate the eigenvalues ​​and the eigenvectors corresponding to the eigenvalues ​​according to the covariance matrix, and sort the eigenvalues ​​in descending order to obtain λ 1 ,λ 2 …λ m , and the eigenvector a corresponding to the eigenvalue 1 , a 2 …a m , and the i-th principal component is Y i =Xa i (1≤i≤m), where X is the original data matrix, a i is the i-th feature vector, and then executes step S33;

[0016] Step S33, based on the eigenvalues ​​arranged in descending order, calculate the principal component contribution rate and the cumulative contribution rate according to the following formula. When the cumulative contribution rate reaches the preset threshold, select the first k principal components to form a new feature space, and then execute step S34:

[0017]

[0018] Among them, λ k is the eigenvalue of the kth principal component, is the sum of all eigenvalues, is the main component contribution rate, is the cumulative contribution rate;

[0019] Step S34, construct a loading matrix based on the eigenvectors of the original data matrix, and obtain the principal component scores of the first k principal components using the following formula:

[0020]

[0021] Among them, PC i represents the principal component score of the i-th principal component, n is the number of total accident numerical features, and w ij is the coefficient of the i-th row and j-th column in the characteristic load matrix, x j is the numerical feature of the jth accident.

[0022] Furthermore, the step S4 uses the elbow method and K-means clustering method according to the following steps to obtain the optimal number of clusters and clustering results:

[0023] Step S41, based on different preset cluster numbers k, calculate the intra-cluster error sum of squares under different cluster numbers k according to the following formula, draw intra-cluster error sum of squares curves under different cluster numbers k, and then execute step S42:

[0024]

[0025] Where WCSS is the intra-cluster error sum of squares, k is the number of clusters, C i represents the i-th cluster, x j is the jth data point belonging to the i-th cluster, c i is the centroid of the i-th cluster;

[0026] Step S42, based on the intra-cluster error square sum curves under different cluster numbers k, select the value at the elbow point of the intra-cluster error square sum curve as the optimal cluster number k, that is, select the k value at the bend of the intra-cluster error square sum curve as the optimal cluster number k;

[0027] Step S43, based on the optimal number of clusters k, randomly initialize k centroid vectors, and calculate the Euclidean distance from each data point to the k centroids according to the following formula, assign each data point to the cluster where the nearest centroid is located, where the data point is the principal component score, and then execute step S44:

[0028]

[0029] Among them, d(x i ,c j ) is the Euclidean distance from each data point to the k centroids, x il is the data point, c jl is the jth centroid, and p is the data dimension;

[0030] Step S44, based on the data points of each cluster, recalculate the centroid vector of each cluster, the centroid position is the mean of all data points in the cluster, and then execute step S45:

[0031]

[0032] Among them, c j is the centroid vector of each cluster, N j is the number of data points in the jth cluster, x i is a data point;

[0033] Step S45, repeating the data point allocation and centroid update steps until the centroid position converges or reaches a predetermined maximum number of iterations.

[0034] Furthermore, the step S5 uses statistical analysis to determine the main features of each category respectively: the accident feature values ​​corresponding to the traffic accident videos contained in each category are counted respectively; if a certain accident feature value accounts for more than 50% in the corresponding category, then the accident feature value is the main feature of the corresponding category; if multiple accident feature values ​​account for more than 50% and 30% in the corresponding category, then the accident feature value is the main feature of the corresponding category; if all accident feature values ​​account for less than 50% in the corresponding category, then the two accident feature values ​​with the largest proportion are selected as the main features of the corresponding category.

[0035] A traffic dangerous driving scene intelligent recognition and classification system, characterized by comprising:

[0036] Data acquisition module: used to obtain a certain number of traffic accident videos;

[0037] Feature extraction module: used to extract accident feature values ​​corresponding to preset accident features in traffic accident videos using deep learning and computer vision technology;

[0038] Dimensionality reduction processing module: used to reduce the dimension of the extracted features based on classification principal component analysis;

[0039] Cluster analysis module: used to classify traffic accident videos into multiple different categories using the elbow method and K-Means clustering method;

[0040] Statistical analysis module: used to determine the main characteristics of each category and the dangerous driving category corresponding to each category.

[0041] Compared with the prior art, the beneficial effects of the present invention are: the accident numerical features of traffic accident videos are reduced in dimension through classification principal component analysis, and the traffic accident videos are divided into multiple categories using cluster analysis, and the main features of each category are determined using statistical analysis, thereby determining the dangerous driving category corresponding to each category; the classification results of the present invention can also be used as input for advanced driver assistance systems (ADAS) and autonomous driving vehicles to assist autonomous driving systems in identifying and making decisions on dangerous driving scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flow chart of a method for intelligently identifying and classifying traffic-hazardous driving scenarios of the present invention.

[0043] Figure 2 It is a schematic diagram of the WCSS curve of the elbow method. DETAILED DESCRIPTION

[0044] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] refer to Figure 1 , a method for intelligently identifying and classifying dangerous traffic driving scenes, characterized in that: for at least two traffic accident videos, the classification of the traffic accident videos is implemented according to steps S1 to S5:

[0046] Step S1, using the driving recorders carried by the accident participants, 1013 traffic accident videos including urban main roads, secondary roads, rural roads and highway ramp sections are obtained, and the accident feature values ​​corresponding to the preset accident features in each traffic accident video are extracted using deep learning and computer vision technology. Specifically, the accident features are the type of the accident participants, the starting and ending positions, the driving trajectory, the movement direction, and the movement distance, and then step S2 is executed;

[0047] Step S2, for each accident feature value corresponding to each traffic accident video, in combination with the optimal scaling rule, each accident feature value is converted into a numerical form, i.e., an accident numerical feature, and the accident numerical features are standardized, i.e., each accident numerical feature is scaled to a range of a mean of 0 and a standard deviation of 1, so as to eliminate the differences between the accident numerical features, and then execute step S3;

[0048] Step S3, based on the accident numerical features corresponding to each traffic accident video, use classification principal component analysis to perform dimensionality reduction to obtain the principal component score, and then execute step S4;

[0049] Specifically, step S3 includes:

[0050] Step S31, constructing an original data matrix based on the accident numerical features corresponding to each traffic accident video, and calculating the covariance matrix of the original data matrix, and then executing step S32;

[0051] Step S32, calculate the eigenvalues ​​and the eigenvectors corresponding to the eigenvalues ​​according to the covariance matrix, and sort the eigenvalues ​​in descending order to obtain λ 1 ,λ 2 …λ m , and the eigenvector a corresponding to the eigenvalue 1 , a 2 …a m , and the i-th principal component is Y i =Xa i (1≤i≤m), where X is the original data matrix, a i is the i-th feature vector, and then executes step S33;

[0052] Step S33, based on the eigenvalues ​​arranged in descending order, the principal component contribution rate and the cumulative contribution rate are calculated according to the following formula. When the cumulative contribution rate reaches 80%, the first k principal components are selected to form a new feature space. In this embodiment, as shown in Table 1, the first three principal components are selected to form a new feature space, and then step S34 is executed:

[0053]

[0054] Among them, λ k is the eigenvalue of the kth principal component, is the sum of all eigenvalues, is the main component contribution rate, is the cumulative contribution rate;

[0055] Table 1

[0056] principal component Eigenvalue Principal component contribution rate (%) 1 1.787 35.741 2 1.467 29.337 3 0.89 17.803 total 4.144 82.881

[0057] Step S34, construct a loading matrix based on the eigenvectors of the original data matrix, and obtain the principal component scores of the first k principal components using the following formula:

[0058]

[0059] Among them, PC i represents the principal component score of the i-th principal component, n is the number of total accident numerical features, and w ij is the coefficient of the i-th row and j-th column in the characteristic load matrix, x j is the numerical feature of the jth accident;

[0060] In this embodiment, the three principal component scores are:

[0061] RC1:y 1 =-0.036x 1 -0.665x 2 +0.690x 3 -0.097x 4 +0.266x 5

[0062] RC2:y 2 =0.629x 1 +0.214x 2 -0.050x 3 -0.470x 4 +0.579x 5

[0063] RC3:y 3 =0.427x 1 -0.128x 2 -0.085x 3 +0.850x 4 +0.267x 5 ;

[0064] Step S4, based on the principal component scores, using the elbow method and K-means clustering method, obtain the optimal number of clusters and clustering results, and divide all traffic accident videos into multiple different categories according to the clustering results, and then execute step S5;

[0065] Specifically, step S4 includes:

[0066] Step S41, based on different preset cluster numbers k, calculate the intra-cluster error sum of squares under different cluster numbers k according to the following formula, draw intra-cluster error sum of squares curves under different cluster numbers k, and then execute step S42:

[0067]

[0068] Where WCSS is the intra-cluster error sum of squares, k is the number of clusters, C i represents the i-th cluster, x j is the jth data point belonging to the i-th cluster, c i is the centroid of the i-th cluster;

[0069] Step S42, based on the intra-cluster error square sum curve under different cluster numbers k, select the value at the elbow point of the intra-cluster error square sum curve as the optimal cluster number k, that is, select the k value at the bend of the intra-cluster error square sum curve as the optimal cluster number k, such as Figure 2 As shown, in this embodiment, the optimal number of clusters is 4;

[0070] Step S43, based on the optimal number of clusters k, randomly initialize k centroid vectors, and calculate the Euclidean distance from each data point to the k centroids according to the following formula, assign each data point to the cluster where the nearest centroid is located, where the data point is the principal component score, and then execute step S44:

[0071]

[0072] Among them, d(x i ,c j ) is the Euclidean distance from each data point to the k centroids, x il is the data point, c jl is the jth centroid, and p is the data dimension;

[0073] Step S44, based on the data points of each cluster, recalculate the centroid vector of each cluster, the centroid position is the mean of all data points in the cluster, and then execute step S45:

[0074]

[0075] Among them, c j is the centroid vector of each cluster, N j is the number of data points in the jth cluster, x i is a data point;

[0076] Step S45, repeating the data point allocation and centroid update steps until the centroid position converges or reaches a predetermined maximum number of iterations;

[0077] Step S5, based on the traffic accident videos included in each category, combined with the accident feature values ​​corresponding to each traffic accident video in each category, use statistical analysis to determine the main features of each category, and then determine the dangerous driving category corresponding to each category, that is, count the accident feature values ​​corresponding to the traffic accident videos included in each category, if a certain accident feature value accounts for more than 50% in the corresponding category, then the accident feature value is the main feature of the corresponding category; if multiple accident feature values ​​account for more than 50% and 30% in the corresponding category, then the accident feature value is the main feature of the corresponding category; if all accident feature values ​​account for less than 50% in the corresponding category, then the two accident feature values ​​with the largest proportion are selected as the main features of the corresponding category; and the determined main features can be used as input to the scene recognition module of the automatic driving system.

[0078] A traffic dangerous driving scene intelligent recognition and classification system, characterized by comprising:

[0079] Data acquisition module: used to obtain a certain number of traffic accident videos;

[0080] Feature extraction module: used to extract accident feature values ​​corresponding to preset accident features in traffic accident videos using deep learning and computer vision technology;

[0081] Dimensionality reduction processing module: used to reduce the dimension of the extracted features based on classification principal component analysis;

[0082] Cluster analysis module: used to classify traffic accident videos into multiple different categories using the elbow method and K-Means clustering method;

[0083] Statistical analysis module: used to determine the main characteristics of each category and the dangerous driving category corresponding to each category.

[0084] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for intelligent identification and classification of dangerous traffic driving scenes, characterized by: For at least two traffic accident videos, the traffic accident videos are classified according to steps S1 to S5: Step S1, obtaining a preset number of traffic accident videos, extracting accident feature values ​​corresponding to preset accident features in each traffic accident video using deep learning and computer vision technology, and then executing step S2; Step S2, for each accident feature value corresponding to each traffic accident video, in combination with a preset conversion rule, each accident feature value is converted into a numerical form, that is, an accident numerical feature, and the accident numerical feature is standardized, and then step S3 is executed; Step S3, based on the accident numerical features corresponding to each traffic accident video, use classification principal component analysis to perform dimensionality reduction to obtain the principal component score, and then execute step S4; Step S4, based on the principal component scores, using the elbow method and K-means clustering method, obtain the optimal number of clusters and clustering results, and divide all traffic accident videos into multiple different categories according to the clustering results, and then execute step S5; Step S5, based on the traffic accident videos included in each category and in combination with the accident feature values ​​corresponding to each traffic accident video in each category, statistical analysis is used to determine the main features of each category, and then determine the dangerous driving category corresponding to each category.

2. The method for intelligent identification and classification of traffic-hazardous driving scenes according to claim 1 is characterized in that: The step S1 acquires the traffic accident video from the image capture device carried by the accident participant or non-accident participant.

3. The method for intelligent identification and classification of traffic-hazardous driving scenes according to claim 1 is characterized in that: The step S3 is based on the following steps, using classification principal component analysis to perform dimensionality reduction and obtain principal component scores: Step S31, constructing an original data matrix based on the accident numerical features corresponding to each traffic accident video, and calculating the covariance matrix of the original data matrix, and then executing step S32; Step S32, calculate the eigenvalues ​​and the eigenvectors corresponding to the eigenvalues ​​according to the covariance matrix, and sort the eigenvalues ​​in descending order to obtain λ1, λ2, ...λ m , and the eigenvectors a1, a2…a corresponding to the eigenvalues m , and the i-th principal component is Y i =Xa i (1≤i≤m), where X is the original data matrix, a i is the i-th feature vector, and then executes step S33; Step S33, based on the eigenvalues ​​arranged in descending order, calculate the principal component contribution rate and the cumulative contribution rate according to the following formula. When the cumulative contribution rate reaches a preset threshold, select the first k principal components to form a new feature space, and then execute step S34: Among them, λ k is the eigenvalue of the kth principal component, is the sum of all eigenvalues, is the main component contribution rate, is the cumulative contribution rate; Step S34, construct a loading matrix based on the eigenvectors of the original data matrix, and obtain the principal component scores of the first k principal components using the following formula: Among them, PC i represents the principal component score of the ith principal component, n is the number of total accident numerical features, and w ij is the coefficient of the i-th row and j-th column in the characteristic load matrix, x j is the numerical feature of the jth accident.

4. The method for intelligent identification and classification of traffic-hazardous driving scenes according to claim 1 is characterized in that: The step S4 uses the elbow method and K-means clustering method according to the following steps to obtain the optimal number of clusters and clustering results: Step S41, based on different preset cluster numbers k, calculate the intra-cluster error sum of squares under different cluster numbers k according to the following formula, draw intra-cluster error sum of squares curves under different cluster numbers k, and then execute step S42: Where WCSS is the intra-cluster error sum of squares, k is the number of clusters, C i represents the i-th cluster, x j is the jth data point belonging to the i-th cluster, c i is the centroid of the i-th cluster; Step S42, based on the intra-cluster error square sum curves under different cluster numbers k, select the value at the elbow point of the intra-cluster error square sum curve as the optimal cluster number k, that is, select the k value at the bend of the intra-cluster error square sum curve as the optimal cluster number k; Step S43, based on the optimal number of clusters k, randomly initialize k centroid vectors, and calculate the Euclidean distance from each data point to the k centroids according to the following formula, assign each data point to the cluster where the nearest centroid is located, where the data point is the principal component score, and then execute step S44: Among them, d(x i ,c j ) is the Euclidean distance from each data point to the k centroids, x il is the data point, c jl is the jth centroid, and p is the data dimension; Step S44, based on the data points of each cluster, recalculate the centroid vector of each cluster, the centroid position is the mean of all data points in the cluster, and then execute step S45: Among them, c j is the centroid vector of each cluster, N j is the number of data points in the jth cluster, x i is a data point; Step S45, repeating the data point allocation and centroid update steps until the centroid position converges or reaches a predetermined maximum number of iterations.

5. The method for intelligent identification and classification of traffic-hazardous driving scenes according to claim 1 is characterized in that: The step S5 uses statistical analysis to determine the main features of each category, specifically: the accident feature values ​​corresponding to the traffic accident videos contained in each category are counted respectively, and if a certain accident feature value accounts for more than 50% in the corresponding category, then the accident feature value is the main feature of the corresponding category; If multiple accident characteristic values ​​account for more than 50% and 30% in the corresponding category, the accident characteristic value is the main characteristic of the corresponding category; If the proportion of all accident characteristic values ​​in the corresponding category does not exceed 50%, the two accident characteristic values ​​with the largest proportion are selected as the main characteristics of the corresponding category.

6. An intelligent recognition and classification system for traffic hazardous driving scenarios, characterized in that: include: Data acquisition module: used to obtain a certain number of traffic accident videos; Feature extraction module: used to extract accident feature values ​​corresponding to preset accident features in traffic accident videos using deep learning and computer vision technology; Dimensionality reduction processing module: used to reduce the dimension of the extracted features based on classification principal component analysis; Cluster analysis module: used to classify traffic accident videos into multiple different categories using the elbow method and K-Means clustering method; Statistical analysis module: used to determine the main characteristics of each category and the dangerous driving category corresponding to each category.

Images