Non-motor vehicle conflict interaction style classification identification method and system

Abstract

The invention relates to a non-motor vehicle conflict interaction style classification identification method and system. The method comprises the following steps: extracting vehicle and non-motor vehicle tracks from an intersection traffic flow video; screening a track fragment combination in which tracks of non-motor vehicles and other traffic participants intersect in the intersection, and taking the track fragment combination as a candidate sample; for each candidate sample, acquiring motion state parameters of the non-motor vehicle and other traffic participants before reaching the track intersection; according to the motion state parameter of each candidate sample, screening samples conforming to a conflict interaction scene; clustering samples obtained by screening by taking the motion state parameters of the non-motor vehicles as characteristics to obtain different types of interaction style characteristics; and constructing an interactive style feature library, performing similarity matching on the non-motor vehicle motion feature index to be identified and the interactive style feature library, and outputting an interactive style identification result. Compared with the prior art, multi-dimensional interaction style recognition can be effectively carried out on the non-motor vehicle in a complex interaction environment.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation, and particularly to a classification and recognition method and system for non-motor vehicle conflict interaction styles. Background Art

[0002] In the current urban traffic system, there are a large number of non-motor vehicles, such as bicycles, two-wheeled electric vehicles, etc. These non-motor vehicles account for a quite large proportion in the traffic flow in the city center and bustling areas. Especially at intersections, due to high traffic density and complex road structures, the conflict interactions between non-motor vehicles and cars are frequent, making them important objects in traffic management. Compared with cars, non-motor vehicles have higher flexibility and mobility, so their driving behaviors are often more variable and difficult to predict. Such variable driving behaviors form diverse interaction styles in the traffic system, especially for non-motor vehicles, which often have more complex behavioral characteristics.

[0003] However, the current recognition and classification system for non-motor vehicle driving styles is not yet perfect, lacking a multi-dimensional and systematic framework to comprehensively evaluate and analyze non-motor vehicle driving behaviors. Most traditional traffic behavior classification methods are designed for cars. Although these methods have achieved remarkable application effects in car driving style analysis, due to significant differences between non-motor vehicles and cars in mechanical structures, motion characteristics, driving behaviors, etc., the existing characteristic parameters and classification results usually cannot be directly transferred to the recognition of non-motor vehicle interaction styles.

[0004] In addition, existing research only models the driving behaviors of non-motor vehicles. For example, Chinese Patent CN118445990A defines the attention range of non-motor vehicles and the effective range of the lateral component force of non-motor vehicles to achieve multi-modal intention prediction. Its essence is to realize two different modal driving behaviors of avoidance and overtaking, and effectively construct long-tail scenarios, but it cannot effectively and accurately identify and classify the interaction styles of non-motor vehicles in complex interaction environments. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above-mentioned defects existing in the prior art and provide a classification and recognition method and system for non-motor vehicle conflict interaction styles, which can realize multi-dimensional interaction style recognition in complex interaction environments and break through the limitations of traditional style classification methods in the non-motor vehicle field.

[0006] The purpose of the present invention can be achieved through the following technical solutions: A classification and recognition method for non-motor vehicle conflict interaction styles, comprising the following steps:

[0007] S1. Obtain the intersection traffic flow video and extract the trajectories of cars and non-motor vehicles therefrom;

[0008] S2. Screen the combinations of trajectory segments where the trajectories of non-motor vehicles and other traffic participants intersect at intersections. Each combination constitutes a candidate sample.

[0009] S3. For each candidate sample, obtain the motion state parameters of non-motor vehicles and other traffic participants before reaching the trajectory intersection point.

[0010] S4. According to the motion state parameters of each candidate sample, screen out the samples that meet the conflict interaction scenarios.

[0011] S5. Using the motion state parameters of non-motor vehicles as features, adopt a clustering algorithm to perform clustering analysis on the samples screened in step S4, complete the classification of non-motor vehicle interaction styles, and obtain the interaction style features of different categories.

[0012] S6. Based on the interaction style features of different categories, construct an interaction style feature library, perform similarity matching between the motion feature indicators of the non-motor vehicle to be recognized and the interaction style feature library, and output the corresponding interaction style recognition results.

[0013] Further, step S1 includes the following steps:

[0014] S11. Shoot and obtain the intersection traffic flow video from a top-down perspective.

[0015] S12. Use an improved model based on the YOLO series to perform object detection on the intersection traffic flow video, and at the same time use a trajectory tracking algorithm based on the SORT framework to extract trajectories, obtaining the trajectories of cars and non-motor vehicles.

[0016] Further, the other traffic participants include cars and other non-motor vehicles.

[0017] Further, the motion state parameters in step S3 include speed, acceleration, deceleration, and yaw rate.

[0018] Further, step S4 includes the following steps:

[0019] S41. Calculate the post-encroachment time PET of the trajectories of non-motor vehicles and other traffic participants, and set the PET threshold according to the intersection scale.

[0020] S42. Calculate the maximum values of the acceleration, deceleration, and yaw rate of non-motor vehicles, and select the lower quartile of all candidate samples as the threshold for this feature parameter.

[0021] S43. When the absolute value of PET is less than the PET threshold, or any one of the absolute values of the acceleration, deceleration, and yaw rate of non-motor vehicles is greater than the corresponding threshold, classify this candidate sample as a conflict interaction scenario; otherwise, discard this candidate sample.

[0022] Further, step S5 includes the following steps:

[0023] S51. Perform principal component analysis dimensionality reduction on the features to be clustered, and retain the feature dimensions with a cumulative variance contribution rate greater than or equal to the preset contribution rate threshold;

[0024] S52. Select multiple numbers of clusters based on the clustering algorithm, and perform clustering analysis on the samples in turn;

[0025] S53. Evaluate the clustering results of each number of clusters according to the clustering result evaluation index, and take the number of clusters with a number of clusters less than the set number and an excellent evaluation result as the final clustering result, that is, obtain the interaction style features corresponding to different categories.

[0026] Further, step S6 specifically calculates the Mahalanobis distance between the non-motor vehicle motion feature index to be recognized and the interaction style features of each category in the interaction style feature library, and takes the interaction style category corresponding to the minimum Mahalanobis distance as the preliminary recognition result;

[0027] Compare the minimum Mahalanobis distance with the matching threshold of the corresponding interaction style category to determine the final recognition result.

[0028] Further, the calculation formula of the Mahalanobis distance is:

[0029]

[0030] where \(x\) is the non-motor vehicle motion feature index to be recognized, \(\mu\) is the mean of each index in a certain category in the interaction style feature library, and \(\Sigma\) is the inverse covariance matrix of the category index.

[0031] Further, the calculation formula of the matching threshold is:

[0032] D threshold =\(\mu + 2\sigma\)

[0033] where \(\mu\) is the average Mahalanobis distance between samples within the class, and \(\sigma\) is the standard deviation between samples within the class;

[0034] When the minimum Mahalanobis distance \(D\) min \(\leq D\) threshold it indicates that it belongs to the interaction style of this category;

[0035] When the minimum Mahalanobis distance \(D\) min \(> D\) threshold it indicates that it does not belong to the interaction style of any known category and is regarded as an abnormal category.

[0036] A non-motor vehicle conflict interaction style classification and recognition system, comprising a data acquisition module, a trajectory extraction module, a feature index calculation module, an interaction style classification module, and an interaction style recognition module. The data acquisition module is used to collect traffic flow video data of non-motor vehicles and automobiles driving at intersections;

[0037] The trajectory extraction module is used to extract the driving trajectory data of non-motor vehicles and automobiles from the video data;

[0038] The feature index calculation module is used to screen out samples that meet the conflict interaction scenario from the trajectory dataset and calculate the motion parameter indexes of non-motor vehicles and automobiles in the samples;

[0039] The interaction style classification module is used to perform cluster analysis on the feature indexes, select the number of clusters according to the evaluation indexes of the clustering results, complete the classification of the interaction style, and obtain the interaction style features of different categories;

[0040] The interaction style recognition module is used to analyze the motion feature indexes of the non-motor vehicle to be recognized, and combine the interaction style features of different categories to determine the specific interaction style of the non-motor vehicle.

[0041] Compared with the prior art, the present invention has the following advantages:

[0042] The present invention first screens out the trajectory segment combinations where the trajectories of non-motor vehicles and other traffic participants intersect as candidate samples; then for each candidate sample, obtains the motion state parameters of the non-motor vehicle and other traffic participants before reaching the trajectory intersection point; then according to the motion state parameters of each candidate sample, screens out the samples that meet the conflict interaction scenario, and uses the motion state parameters of the non-motor vehicle as features, and adopts a clustering algorithm to perform cluster analysis on the samples that meet the conflict interaction scenario to obtain the interaction style features of different categories, which are used to construct an interaction style feature library. Subsequently, the motion feature indexes of the non-motor vehicle to be recognized are matched with the interaction style feature library for similarity, and the corresponding interaction style recognition result can be determined. Thus, it can realize multi-dimensional interaction style recognition in a complex interaction environment, breaking through the limitations of traditional style classification methods in the field of non-motor vehicles.

[0043] The present invention calculates multiple motion parameters of non-motor vehicles and other traffic participants (cars or non-motor vehicles) before reaching the trajectory intersection in the intersection, including speed, acceleration, deceleration, yaw rate of angular velocity, etc. On the one hand, samples that meet the conflict interaction scenario are screened out using the motion state parameters. On the other hand, the motion state parameters are used as features for clustering analysis to obtain different interaction style features. By combining the interaction features of non-motor vehicles and cars, a comprehensive and detailed multi-dimensional classification of the driving styles of non-motor vehicles is achieved. This multi-dimensional analysis can not only capture the behavioral characteristics of non-motor vehicles more accurately, but also reveal the reasons for the behavioral changes of non-motor vehicles in complex interaction environments, providing richer data support for traffic management.

[0044] Considering the obvious fluctuation of the yaw rate of angular velocity unique to non-motor vehicles, the present invention selects the yaw rate of angular velocity of non-motor vehicles as one of the clustering features, breaking through the limitations of traditional car driving style classification methods in the field of non-motor vehicles. By targeting the unique dynamic behaviors and motion characteristic parameters of non-motor vehicles, the present invention can better adapt to severe conflict interaction scenarios in complex road traffic environments, improving the accuracy and robustness of interaction style classification.

[0045] When the present invention uses a clustering algorithm to perform clustering analysis on samples that meet the conflict interaction scenario, it first performs principal component analysis dimensionality reduction on the features to be clustered, retaining the feature dimensions with a cumulative variance contribution rate greater than or equal to the preset contribution rate threshold, which can effectively improve the clustering effect and calculation efficiency. In addition, according to the clustering result evaluation index, the clustering results of the number of clusters in each category are evaluated, and the number of clusters with a small number of clusters and excellent evaluation results is selected as the final clustering result, which can ensure accurate classification to obtain different interaction style features. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a schematic flow chart of the method of the present invention;

[0047] Figure 2 is a schematic structural diagram of the system of the present invention;

[0048] Figure 3 is a change diagram of the evaluation index under different numbers of clusters in the embodiment;

[0049] Figures 4a - 4j is a statistical chart of the interaction style features of each category when the number of clusters is 3 in the embodiment;

[0050] Explanation of marks in the figure: 1. Data acquisition module, 2. Trajectory extraction module, 3. Feature index calculation module, 4. Interaction style classification module, 5. Interaction style recognition module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] Embodiment

[0053] As Figure 1 shown, a method for classifying and recognizing non-motor vehicle conflict interaction styles includes the following steps:

[0054] S1. Obtain the intersection traffic flow video and extract the trajectories of cars and non-motor vehicles from it;

[0055] S2. Screen the trajectory segment combinations where the trajectories of non-motor vehicles and other traffic participants intersect in the intersection. Among them, each combination constitutes a candidate sample;

[0056] S3. For each candidate sample, obtain the motion state parameters of the non-motor vehicle and other traffic participants before reaching the trajectory intersection point;

[0057] S4. According to the motion state parameters of each candidate sample, screen out the samples that meet the conflict interaction scenario;

[0058] S5. Using the motion state parameters of the non-motor vehicle as features, adopt a clustering algorithm to perform clustering analysis on the samples screened in step S4, complete the classification of non-motor vehicle interaction styles, and obtain the interaction style features of different categories;

[0059] S6. Based on the interaction style features of different categories, construct an interaction style feature library, perform similarity matching between the motion feature indicators of the non-motor vehicle to be recognized and the interaction style feature library, and output the corresponding interaction style recognition result.

[0060] Based on the above method, a non-motor vehicle conflict interaction style classification and recognition system is implemented. As Figure 2 shown, it includes a data acquisition module 1, a trajectory extraction module 2, a feature index calculation module 3, an interaction style classification module 4, and an interaction style recognition module 5 connected in sequence. Among them, the data acquisition module 1 is used to collect video data of non-motor vehicles and cars driving at the intersection;

[0061] The trajectory extraction module 2 is used to extract the driving trajectory data of non-motor vehicles and cars from the video data;

[0062] The feature index calculation module 3 is used to determine the samples of conflict interaction from the trajectory dataset and calculate various motion parameter indicators of non-motor vehicles and cars in the samples;

[0063] The interaction style classification module 4 is used to perform clustering analysis on the feature indexes, and select a reasonable number of clusters according to the clustering result evaluation indexes to complete the classification of the interaction style;

[0064] The interaction style recognition module 5 is used to analyze the input non-motor vehicle movement feature indicators and determine the specific interaction style of the non-motor vehicle according to the interaction style category features.

[0065] This embodiment applies the above scheme, and the main contents are as follows:

[0066] First, select intersections with a high density of non-motor vehicles, and use a drone to shoot the traffic flow video from a top-down perspective, and extract the trajectories of cars and non-motor vehicles from the traffic flow video; in this embodiment, the YOLOv8 object detection algorithm is used to perform high-precision object detection on the traffic flow video, and at the same time, the Deep-SORT trajectory tracking algorithm is used to complete trajectory extraction.

[0067] After that, screen the trajectory segment combinations where the trajectories of non-motor vehicles intersect with cars or other non-motor vehicles in the intersection. Each combination constitutes a candidate sample;

[0068] Then, calculate the motion state parameters of non-motor vehicles and cars or other non-motor vehicles before reaching the trajectory intersection point in each candidate sample. Among them, the motion state parameters include average speed, maximum speed, average acceleration, maximum acceleration, average deceleration, maximum deceleration, average yaw rate of angular velocity, maximum yaw rate of angular velocity, and standard deviation of yaw rate of angular velocity.

[0069] According to the motion state parameters of each candidate sample, screen out the samples that meet the conflict interaction scenario. Specifically:

[0070] On the one hand, calculate the post-encroachment time PET of the non-motor vehicle and car or other non-motor vehicle trajectories. In this embodiment, the PET threshold is set to 5 s according to the intersection scale;

[0071] On the other hand, calculate the maximum values of the acceleration, deceleration, and yaw rate of angular velocity of non-motor vehicles, and select the lower quartile of all statistical samples as the corresponding parameter threshold for this feature parameter;

[0072] When the absolute value of PET is less than the PET threshold, or any one of the absolute values of the acceleration, deceleration, and yaw rate of angular velocity of the non-motor vehicle is greater than the corresponding parameter threshold, classify the candidate sample as a conflict interaction scenario between the non-motor vehicle and the car or other non-motor vehicle, otherwise discard the candidate sample.

[0073] This embodiment finally screens and obtains 1384 groups of conflict interaction samples of straight two-wheelers as samples to be clustered and analyzed.

[0074] Taking the motion state parameters of non-motor vehicles as features, use a clustering algorithm to perform clustering analysis on the samples to complete the classification of non-motor vehicle interaction styles and obtain different category interaction style features. Specifically:

[0075] Perform principal component analysis dimensionality reduction on the features to be clustered. In this embodiment, the feature dimensions with a cumulative variance contribution rate ≥ 85% are retained to improve the clustering effect and calculation efficiency;

[0076] In this embodiment, K-means++ is selected as the clustering algorithm to process and analyze the motion feature information. The number of clusters is selected from 2 to 10, and the samples are clustered and analyzed in turn. It should be noted that in practical applications, other clustering algorithms such as the DBSCAN clustering algorithm can also be used to process and analyze the motion feature information to identify the driving styles of different non-motor vehicles during the interaction process.

[0077] In this embodiment, the Silhouette Score (SS) and the Davies-Bouldin index (DBI) are selected as the clustering result evaluation indicators to evaluate the clustering results of different numbers of clusters. The number of clusters with fewer clusters and better evaluation results is comprehensively selected as the final clustering result. Figure 3 The results of the two clustering evaluation indicators under 2-10 clusters are shown. Among them, the higher the SS (left coordinate axis) and the lower the DBI (right coordinate axis), the better the clustering result. Based on the results of the two indicators, the final number of clusters selected in this embodiment is 3, Figures 4a - 4j The overall results of each feature under 3 clusters are correspondingly shown.

[0078] Based on the interaction style features of different categories, an interaction style feature library is constructed, and the motion feature indicators of the non-motor vehicle to be recognized are matched with the interaction style features of each category, and the corresponding interaction style category is output.

[0079] In this embodiment, when performing the similarity matching process, first calculate the Mahalanobis distance between the motion feature indicators of the non-motor vehicle to be recognized and each interaction style category in the feature library:

[0080]

[0081] where x is the motion feature indicator of the non-motor vehicle to be recognized, μ is the mean of each indicator in a certain category of interaction style, and Σ is the inverse covariance matrix of the indicators of this category.

[0082] After calculating the Mahalanobis distance of the motion feature indicators of the non-motor vehicle to be recognized relative to each category, select the category corresponding to the minimum Mahalanobis distance as the preliminary recognition result, and then compare it with the set matching threshold:

[0083] D threshokd = μ + 2σ

[0084] where μ is the average Mahalanobis distance between samples within the class, and σ is the standard deviation between samples within the class.

[0085] When the minimum Mahalanobis distance D min ≤D threshokd , it is considered that the non-motor vehicle to be recognized belongs to this interaction style category; when the minimum Mahalanobis distance D min >D threshold , it is considered that the non-motor vehicle to be recognized does not belong to any known interaction style category and is regarded as an abnormal category.

[0086] In this embodiment, the motion characteristic indexes of the non-motor vehicle to be recognized are:

[0087] x = [-2.5, 1.5, 4.5, 0.03, 0.19, 0.02, 0.34, 0.88, -0.6, -0.85] T

[0088] After calculation, it is found that its Mahalanobis distance from the 0th class in Figures 4a - 4j is the closest. Therefore, after being recognized by this step, the interaction style of the non-motor vehicle is determined to be the 0th class.

[0089] In summary, the present solution can accurately recognize and classify the interaction styles of non-motor vehicles. On the one hand, it is convenient for traffic management departments to effectively recognize and monitor the driving behaviors of non-motor vehicles, realizing more refined traffic management; on the other hand, this classification method and result can provide an important basis for the construction of non-motor vehicle simulation models, helping to more accurately simulate the behaviors of non-motor vehicles in complex traffic environments, and providing support for aspects such as the optimization of intelligent transportation systems and the improvement of non-motor vehicle safety.

Claims

1. A method for classifying and identifying non-motor vehicle conflict interaction styles, characterized in that It includes the following steps: S1. Obtain the intersection traffic flow video and extract the trajectories of cars and non-motor vehicles therefrom; S2. Screen the trajectory segment combinations where the trajectories of non-motor vehicles and other traffic participants cross in the intersection. Each combination constitutes a candidate sample; S3. For each candidate sample, obtain the motion state parameters of the non-motor vehicle and other traffic participants before reaching the trajectory intersection point; S4. According to the motion state parameters of each candidate sample, screen out the samples that meet the conflict interaction scenario; S5. Using the motion state parameters of the non-motor vehicle as features, adopt a clustering algorithm to perform clustering analysis on the samples screened in step S4, complete the classification of the non-motor vehicle interaction style, and obtain the interaction style features of different categories; S6. Based on the interaction style features of different categories, construct an interaction style feature library, perform similarity matching between the motion feature indicators of the non-motor vehicle to be recognized and the interaction style feature library, and output the corresponding interaction style recognition result.

2. The non-motor vehicle conflict interaction style classification and recognition method according to claim 1, characterized in that, The step S1 includes the following steps: S11. Shoot and obtain the intersection traffic flow video from a top-down perspective; S12. Use an improved model based on the YOLO series to perform object detection on the intersection traffic flow video, and at the same time adopt a trajectory tracking algorithm based on the SORT framework to extract trajectories, obtaining the trajectories of cars and non-motor vehicles.

3. A non-motor vehicle conflict interaction style classification and recognition method according to claim 1, characterized in that The other traffic participants include cars and other non-motor vehicles.

4. A non-motor vehicle conflict interaction style classification and recognition method according to claim 1, characterized in that The motion state parameters in the step S3 include speed, acceleration, deceleration, and yaw rate of angular velocity.

5. A non-motor vehicle conflict interaction style classification and recognition method according to claim 3, characterized in that The step S4 includes the following steps: S41. Calculate the post-encroachment time PET of the trajectories of non-motor vehicles and other traffic participants, and set the PET threshold according to the intersection scale; S42. Calculate the maximum values of the acceleration, deceleration, and yaw rate of angular velocity of the non-motor vehicle, and select the lower quartile of all candidate samples as the threshold of this feature parameter; S43. When the absolute value of PET is less than the PET threshold, or any one of the absolute values of the acceleration, deceleration, and yaw rate of angular velocity of the non-motor vehicle is greater than the corresponding threshold, classify this candidate sample as a conflict interaction scenario, otherwise discard this candidate sample.

6. The method for classifying and identifying non-motor vehicle conflict interaction styles according to claim 3, characterized in that, The step S5 includes the following steps: S51. Perform principal component analysis and dimensionality reduction processing on the features to be clustered, and retain the feature dimensions with a cumulative variance contribution rate greater than or equal to the preset contribution rate threshold; S52. Based on the clustering algorithm, select multiple numbers of clusters and perform clustering analysis on the samples in turn; S53. According to the clustering result evaluation index, evaluate the clustering results of various numbers of clusters, and take the number of clusters with a number of clusters less than the set number and an evaluation result of excellent as the final clustering result, that is, obtain the interaction style features corresponding to different categories.

7. A method for classifying and identifying non-motor vehicle conflict interaction styles according to claim 6, characterized in that, The step S6 is specifically to calculate the Mahalanobis distance between the motion feature indicators of the non-motor vehicle to be recognized and the interaction style features of each category in the interaction style feature library, and take the interaction style category corresponding to the minimum Mahalanobis distance as the preliminary recognition result; Compare the minimum Mahalanobis distance with the matching threshold of the corresponding interaction style category to determine the final recognition result.

8. A non-motor vehicle conflict interaction style classification and recognition method according to claim 7, characterized in that The calculation formula of the Mahalanobis distance is: Among them, x is the non-motor vehicle movement feature index to be recognized, μ is the mean value of each index in a certain category in the interaction style feature library, and Σ is the inverse covariance matrix of the indexes of this category.

9. A method for classifying and identifying non-motor vehicle conflict interaction styles according to claim 8, characterized in that, The calculation formula of the matching threshold is as follows: D threshold = μ + 2σ Among them, μ is the average Mahalanobis distance between samples within a class, and σ is the standard deviation between samples within a class; When the minimum Mahalanobis distance D min ≤D threshold then it indicates that it belongs to this category of interaction style; When the minimum Mahalanobis distance D min > D threshold then it indicates that it does not belong to any known category interaction style and is regarded as an abnormal category.

10. A non-motor vehicle conflict interaction style classification and recognition system, implemented based on a non-motor vehicle conflict interaction style classification and recognition method as described in claim 1, characterized in that, It includes a data acquisition module, a trajectory extraction module, a feature index calculation module, an interaction style classification module, and an interaction style recognition module. The data acquisition module is used to collect traffic flow video data of non-motor vehicles and automobiles driving at intersections; The trajectory extraction module is used to extract the driving trajectory data of non-motor vehicles and automobiles from the video data; The feature index calculation module is used to screen out samples that meet the conflict interaction scenario from the trajectory dataset and calculate the movement parameter indexes of non-motor vehicles and automobiles in the samples; The interaction style classification module is used to perform clustering analysis on the feature indexes, and select the number of clusters according to the evaluation indexes of the clustering results to complete the classification of the interaction style and obtain different categories of interaction style features; The interaction style recognition module is used to analyze the non-motor vehicle movement feature index to be recognized, and combine different categories of interaction style features to determine the specific interaction style of this non-motor vehicle.

Images