A heavy-load vehicle aggressive driving behavior recognition method based on multi-dimensional kinematic features

Abstract

The application discloses a heavy-load vehicle aggressive driving behavior recognition method based on multi-dimensional kinematic characteristics, belongs to the technical field of driving behavior recognition, and comprises the following steps: acquiring high-frequency kinematic data of a heavy-load vehicle, and performing cutting processing on the high-frequency kinematic data, so as to divide a driving track into independent acceleration events and deceleration events; multi-dimensional indexes representing the operation intensity and stability of a driver are extracted, and an initial feature vector of a corresponding event is constructed; principal component analysis is adopted to perform dimension reduction processing on an original feature matrix, so as to obtain a speed selection factor, an acceleration distribution factor and an operation stability factor; the event cutting based on the kinematic law significantly improves the robustness and logical consistency of feature extraction; the hierarchical clustering architecture effectively solves the behavior evaluation nonlinear problem caused by the difference in the speed domain; and the high-order kinematic derivative and the orthogonalization feature space are introduced, so that the driving stability is deeply represented.

Description

Technical Field

[0001] This invention belongs to the technical field of driving behavior recognition, specifically relating to a method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features. Background Technology

[0002] Heavy trucks, due to their high center of gravity, large mass, and high inertia, are more prone to loss of control during rapid acceleration or emergency braking, significantly increasing the risk of traffic accidents. Current methods for identifying aggressive driving behavior have two main limitations: judgment criteria based on instantaneous state thresholds do not consider the temporal continuity of timely actions, easily leading to false alarms and false negatives; while driving pattern matching strategies relying on predefined behavior templates are difficult to adapt to complex road scenarios. Furthermore, existing research generally focuses on single-dimensional feature analysis, lacking in-depth exploration of the correlations between multi-dimensional features, thus limiting the comprehensiveness and accuracy of the identification results. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned shortcomings in the prior art by providing a method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features, thereby solving the problem that existing methods for recognizing aggressive driving behavior of heavy-duty trucks have significant limitations.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features includes the following steps: S1. Acquire high-frequency kinematic data of heavy-duty vehicles, and perform segmentation processing on the high-frequency kinematic data based on the event segmentation algorithm of the rate of change of speed, dividing the driving trajectory into independent acceleration events and deceleration events. S2. In each segmented acceleration and deceleration event, extract multi-dimensional indicators that characterize the driver's operational intensity and stability, and construct the initial feature vector for the corresponding event. S3. Combine the initial feature vectors corresponding to all acceleration and deceleration events to obtain the original feature matrix. Then, use principal component analysis to reduce the dimensionality of the original feature matrix and decouple it to obtain the velocity selection factor, acceleration distribution factor and operation stability factor. S4. Based on the speed selection factor, acceleration distribution factor, and operational stability factor, a two-stage K-Means clustering method is used to classify driving behavior in order to identify aggressive driving behavior.

[0005] Furthermore, S1 specifically includes: High-frequency kinematic data of heavy-duty vehicles is acquired using on-board sensors, and instantaneous speed and time series are extracted from the high-frequency kinematic data. Based on the event segmentation algorithm of speed change rate, a triple speed filtering threshold is set to divide the driving trajectory into independent acceleration events and deceleration events.

[0006] Furthermore, in S1, the triple velocity filtering threshold includes: velocity change Initial speed And the absolute value of acceleration / deceleration is not less than .

[0007] Furthermore, in S2, the multidimensional indicators characterizing the driver's operational intensity and stability include: maximum speed, average speed, maximum longitudinal acceleration, maximum longitudinal deceleration, average acceleration, average deceleration, maximum value of jerkiness, mean of jerkiness, standard deviation of jerkiness, duration of the event, and standard deviation of speed.

[0008] Furthermore, in S3, principal component analysis is used to perform dimensionality reduction and decoupling of the original feature matrix, including: First, Z-score standardization is performed on the original feature matrix to obtain the standardized matrix. Then, the covariance matrix of the standardized matrix is ​​calculated and its eigenvalues ​​and eigenvectors are solved. Based on the eigenvalues, the cumulative variance contribution rate is calculated, and principal components are selected according to the cumulative variance contribution rate ≥ 80%. Finally, the extracted principal components are defined as velocity selection factor, acceleration distribution factor and operational stability factor, respectively.

[0009] Furthermore, in S4, the two-stage K-Means clustering method includes: The first stage of the driving condition classification is based on the principal component of the speed selection factor and uses the K-Means algorithm to classify driving events into two types of driving conditions: low-speed driving and high-speed driving. The second stage of behavior recognition involves using the K-Means algorithm to cluster the principal components of acceleration distribution factors and operational stability factors under both low-speed and high-speed driving conditions to identify different driving behaviors.

[0010] Furthermore, in the second stage of behavior recognition, driving behavior includes smooth driving, moderate driving, and aggressive driving.

[0011] Furthermore, in the second stage of behavior recognition, the objective function of clustering is to minimize the sum of squared errors within each cluster, which is expressed as: In the formula, is the objective function; K is the number of clusters, corresponding to smooth driving, moderate driving, and aggressive driving, respectively; For the first i A cluster, The principal component feature vector of the driving event. For the first i The centroid vector of each cluster.

[0012] Furthermore, in S4, the Pearson correlation coefficient is introduced to calculate the linear correlation between the degree of aggressive driving behavior and the risk event index; when the linear correlation coefficient is significantly positive, it is determined that the identified aggressive driving is significantly positively correlated with the high-risk attribute, thereby confirming the effectiveness of the identification method.

[0013] The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multi-dimensional kinematic features provided by this invention has the following beneficial effects: 1. Event segmentation based on kinematic laws significantly improves the robustness and logical consistency of feature extraction; This invention abandons the coarse processing methods based on fixed time windows or single instantaneous thresholds in traditional approaches, and innovatively adopts a continuity discrimination mechanism based on the instantaneous rate of change of velocity (acceleration). This is achieved by introducing a strict physical threshold (…). , A filter was constructed to effectively suppress high-frequency noise from the sensor and random disturbances in traffic flow, ensuring that the extracted driving events possess a complete "operation-response" closed loop in a physical sense. This kinematic-based segmentation method guarantees the input quality of subsequent pattern recognition from the data source, enabling the recognition results to truly reflect the driver's intentions and the vehicle's dynamic response.

[0014] 2. The hierarchical clustering architecture effectively solves the nonlinearity problem in behavior evaluation caused by differences in the velocity domain; To address the significant heterogeneity of driving behavior risk thresholds for heavy-duty trucks under different driving conditions (low-speed urban congestion vs. high-speed highways), this invention proposes a two-stage hierarchical clustering algorithm based on "driving condition segmentation and behavior recognition." This architecture first achieves adaptive decoupling of driving conditions based on principal components of speed, and then performs refined clustering of behavioral patterns within specific driving condition subspaces. This design overcomes the shortcomings of traditional single-layer models, which suffer from masking aggressive low-speed features or false alarms of normal high-speed operation due to "meaning" across the entire speed domain. This enables adaptive and high-precision identification of driving risks in variable traffic scenarios.

[0015] 3. By introducing higher-order kinematic derivatives and orthogonalizing the feature space, a deep characterization of driving stability is achieved; This invention, in its feature engineering, not only considers conventional speed and acceleration metrics but also introduces the derivative of acceleration with respect to time—Jerk—as a core dimension for measuring driving smoothness. Combined with Principal Component Analysis (PCA), the original high-dimensional feature space is mapped into an orthogonal low-dimensional subspace containing "speed selection," "acceleration distribution," and "operational stability." This process not only eliminates collinearity and redundancy among features but also endows the model with the ability to identify covert aggressive driving behaviors such as "low-speed, high-frequency impacts," significantly improving the sensitivity in identifying potentially risky drivers. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features in the embodiment.

[0017] Figure 2 This is a logic block diagram of the aggressive driving behavior recognition method for heavy-duty vehicles based on multidimensional kinematic features in the embodiment.

[0018] Figure 3 This is an event segmentation diagram from the embodiment.

[0019] Figure 4 This is a loading diagram of the PCA principal component analysis in the example. Detailed Implementation

[0020] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0021] This embodiment presents a method for identifying aggressive driving behavior in heavy-duty vehicles based on multidimensional kinematic features. It identifies aggressive driving behavior through multidimensional kinematic feature analysis and reveals its intrinsic correlation mechanism with accident risk. The study first defines extraction criteria for longitudinal acceleration and braking events, obtaining driving events and their kinematic feature parameters from trajectory data. Principal component analysis is used to reduce the dimensionality of the multidimensional features, categorizing them into three feature sets: speed selection, acceleration distribution, and operational stability. Then, a two-stage clustering analysis method is employed, combining speed and operational features, to analyze aggressive driving behavior under different speed conditions; referencing… Figure 1 and Figure 2 Specifically, it includes the following: S1. Acquire high-frequency kinematic data of heavy-duty vehicles, and perform segmentation processing on the high-frequency kinematic data based on the event segmentation algorithm of the rate of change of speed, dividing the driving trajectory into independent acceleration events and deceleration events. In some embodiments, on-board sensors (such as CAN bus, GPS, and ADAS devices) are used to acquire raw high-frequency kinematic data of heavy-duty vehicles, and instantaneous speed and time series are extracted. In order to extract statistically significant operation segments from continuous driving flow, this invention adopts an event segmentation algorithm based on the rate of change of speed. This algorithm uses the trend of instantaneous speed change as the segmentation basis to divide the complete driving trajectory into independent acceleration events and deceleration events.

[0022] In one specific embodiment, reference is made to Figure 3 During the segmentation process, in order to eliminate sensor noise and subtle speed fluctuations, a strict triple speed filtering threshold was set: only the speed change was retained. Initial speed And the absolute value of acceleration / deceleration is not less than This approach ensures that subsequent data analysis stems from genuine acceleration or braking behavior, providing a clean data foundation for identifying aggressive driving characteristics.

[0023] S2. In each segmented acceleration and deceleration event, extract multi-dimensional indicators that characterize the driver's operational intensity and stability, and construct the initial feature vector for the corresponding event. In some embodiments, within each segmented acceleration or deceleration event, multidimensional indicators that characterize the driver's operational intensity and stability are further extracted. In this embodiment, 11 key indicators are selected to form an initial feature vector, including: maximum speed, average speed, maximum longitudinal acceleration, maximum longitudinal deceleration, average acceleration, average deceleration, maximum value of jerkiness, mean of jerkiness and standard deviation of jerkiness, as well as duration of the event and standard deviation of speed.

[0024] Taking the standard deviation of jerkiness as an example, its calculation formula is: In the formula, For speed, For acceleration, For time; These indicators not only cover drivers' speed preferences (such as high or low speeds), but also delve deeper into the "explosiveness" (reflected by peak acceleration) and "smoothness" (reflected by fluctuations in agility) of drivers during operation. Through this multi-dimensional characterization, the driving habits of aggressive drivers, such as frequently and sharply pressing the accelerator or brake, which involve high-energy impacts, can be fully captured.

[0025] S3. Combine the initial feature vectors corresponding to all acceleration and deceleration events to obtain the original feature matrix. Then, use principal component analysis to reduce the dimensionality of the original feature matrix and decouple it to obtain the velocity selection factor, acceleration distribution factor and operation stability factor. In some embodiments, reference Figure 4 Based on Principal Component Analysis (PCA) for feature dimensionality reduction and decoupling, this invention addresses the issue that the 11 extracted kinematic indices often exhibit strong autocorrelation (e.g., maximum acceleration and average acceleration are usually highly positively correlated), as direct clustering would lead to uneven feature weight distribution. Therefore, this invention introduces Principal Component Analysis (PCA) to reduce the dimensionality of the original feature matrix. Through orthogonal transformation, the high-dimensional features are mapped to an uncorrelated low-dimensional principal component space.

[0026] In one specific embodiment, firstly, the original feature matrix is... Z-score standardization is performed to obtain the standardized matrix z, where any element The calculation formula is: in, For the first The first driving incident The original values ​​of each feature; For the first of all events The mean of each feature; For the first of all events The standard deviation of each feature is then calculated. Next, the covariance matrix C of the standardized matrix is ​​calculated, and its eigenvalues ​​are solved. and eigenvectors. Based on the cumulative variance contribution rate. The standard formula for selecting the first k principal components is as follows: In the formula, The number of principal components selected. Let i be the eigenvalue of the i-th principal component. Let m be the eigenvalue of the m-th principal component. The index variable of the principal component, This represents the total number of principal components.

[0027] Experiments showed that the cumulative variance contribution rate of the first few principal components can usually reach over 80%, effectively representing all the original index information. The finally extracted principal components can be clearly defined as "velocity selection factor," "acceleration distribution factor," and "operational stability factor," which not only simplifies the calculation model but also eliminates the interference of redundant information on recognition accuracy.

[0028] S4. Based on the speed selection factor, acceleration distribution factor and operational stability factor, a two-stage K-Means clustering method is used to classify driving behavior in order to identify aggressive driving behavior. In some embodiments, the two-stage K-Means clustering method includes: The first stage of driving condition classification: Based on the principal component of the speed selection factor, the K-Means algorithm is used to classify driving events into two types of driving conditions: low-speed driving and high-speed driving. In the second stage of behavior recognition, within the defined operating condition categories, the K-Means algorithm is further used to cluster the principal components of "acceleration distribution" and "operational stability" to identify driving modes (behaviors) with different degrees of aggression: smooth driving, moderate driving, and aggressive driving.

[0029] Among them, the objective function of clustering To minimize the sum of squared errors within the cluster: In the formula, is the objective function; K is the number of clusters, with a value of 3, corresponding to smooth driving, moderate driving, and aggressive driving, respectively; For the first i A cluster, The principal component feature vector of the driving event. For the first i The centroid vector of each cluster.

[0030] In some embodiments, the silhouette coefficient is used to ensure the validity of the clustering results. The evaluation formula is as follows: In the formula, Let i be the average distance between sample i and the nearest other clusters. Let be the average distance between sample i and other samples in its cluster; In some embodiments, the Pearson correlation coefficient is introduced. The linear correlation between the degree of aggressive driving behavior and risk event indicators is calculated. When the linear correlation coefficient is significantly positive, it is determined that the identified aggressive driving is significantly positively correlated with the high-risk attribute, thus confirming the effectiveness of the identification method. The calculation formula is as follows: In the formula, Let X be the value of the i-th sample. Let X be the mean of the variable. Let Y be the value of the i-th sample. Let Y be the mean of the variable. This represents the number of samples.

[0031] This invention extracts a "speed selection factor" using PCA. In the first stage, it first divides the low-speed / high-speed operating conditions of heavy-duty vehicles, and then performs clustering based on "acceleration distribution factor and operational stability factor" within the corresponding operating conditions. This "operating condition division first, risk clustering later" approach avoids the shortcomings of the small car solution, which "directly judges aggressive driving based solely on acceleration indicators," and achieves accurate risk identification for heavy-duty vehicles under different load-speed combinations, significantly reducing the misjudgment rate.

[0032] This invention presents a multi-level risk verification mechanism that addresses the problem of the ambiguous causal relationship between "aggressive driving and accident risk" in heavy-duty vehicles. It introduces the Pearson correlation coefficient to verify the correlation between identified driving modes (smooth / moderate / aggressive) and specific risk indicators for heavy-duty vehicles (such as the number of emergency braking events, the number of rapid acceleration / deceleration events, and rollover risk scores), rather than relying solely on clustering profile coefficients. This mechanism clearly quantifies the strong positive correlation between "aggressive driving modes of heavy-duty vehicles" and high-risk events, resolving the industry pain point of "inability to verify the validity of identification results" in heavy-duty scenarios, and providing direct data support for subsequent risk warning strategies.

[0033] This invention, based on the high-dimensional driving characteristics of heavy-duty vehicles, extracts three core principal components—"speed selection factor, acceleration distribution factor, and operational stability factor"—through PCA, rather than a general combination of driving features. The "operational stability factor" specifically corresponds to the unique behavioral characteristics of heavy-duty vehicles, such as load fluctuations and vehicle attitude control. This achieves scenario-based and targeted dimensionality reduction of high-dimensional features, reducing computational complexity while retaining key risk information in heavy-duty scenarios, thus adapting to the characteristics of large amounts and high dimensionality of heavy-duty vehicle driving behavior data.

[0034] Although specific embodiments of the invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of this patent. Various modifications and variations that can be made by a person skilled in the art without inventive effort within the scope described in the claims still fall within the scope of protection of this patent.

Claims

1. A method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features, characterized in that, Includes the following steps: S1. Acquire high-frequency kinematic data of heavy-duty vehicles, and perform segmentation processing on the high-frequency kinematic data based on the event segmentation algorithm of the rate of change of speed, dividing the driving trajectory into independent acceleration events and deceleration events. S2. In each segmented acceleration and deceleration event, extract multi-dimensional indicators that characterize the driver's operational intensity and stability, and construct the initial feature vector for the corresponding event. S3. Combine the initial feature vectors corresponding to all acceleration and deceleration events to obtain the original feature matrix. Then, use principal component analysis to reduce the dimensionality of the original feature matrix and decouple it to obtain the velocity selection factor, acceleration distribution factor and operation stability factor. S4. Based on the speed selection factor, acceleration distribution factor, and operational stability factor, a two-stage K-Means clustering method is used to classify driving behavior in order to identify aggressive driving behavior.

2. The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features according to claim 1, characterized in that, S1 specifically includes: High-frequency kinematic data of heavy-duty vehicles is acquired using on-board sensors, and instantaneous speed and time series are extracted from the high-frequency kinematic data. Based on the event segmentation algorithm of speed change rate, a triple speed filtering threshold is set to divide the driving trajectory into independent acceleration events and deceleration events.

3. The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features according to claim 2, characterized in that, In S1, the triple velocity filtering threshold includes: velocity change Initial speed And the absolute value of acceleration / deceleration is not less than .

4. The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features according to claim 1, characterized in that, In S2, the multidimensional indicators characterizing the driver's operational intensity and stability include: maximum speed, average speed, maximum longitudinal acceleration, maximum longitudinal deceleration, average acceleration, average deceleration, maximum value of jerkiness, mean of jerkiness, standard deviation of jerkiness, duration of event, and standard deviation of speed.

5. The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multi-dimensional kinematic features according to claim 1, characterized in that, In step S3, principal component analysis is used to reduce the dimensionality of the original feature matrix and decouple it, including: First, Z-score standardization is performed on the original feature matrix to obtain the standardized matrix. Then, the covariance matrix of the standardized matrix is ​​calculated and its eigenvalues ​​and eigenvectors are solved. Based on the eigenvalues, the cumulative variance contribution rate is calculated, and principal components are selected according to the cumulative variance contribution rate ≥ 80%. Finally, the extracted principal components are defined as velocity selection factor, acceleration distribution factor and operational stability factor, respectively.

6. The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features according to claim 1, characterized in that, In S4, the two-stage K-Means clustering method includes: The first stage of the driving condition classification is based on the principal component of the speed selection factor and uses the K-Means algorithm to classify driving events into two types of driving conditions: low-speed driving and high-speed driving. The second stage of behavior recognition involves using the K-Means algorithm to cluster the principal components of acceleration distribution factors and operational stability factors under both low-speed and high-speed driving conditions to identify different driving behaviors.

7. The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features according to claim 6, characterized in that, In the second stage of behavior recognition, driving behaviors include smooth driving, moderate driving, and aggressive driving.

8. The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features according to claim 6, characterized in that, In the second stage of behavior recognition, the objective function of clustering is to minimize the sum of squared errors within each cluster, which is expressed as: In the formula, is the objective function; K is the number of clusters, corresponding to smooth driving, moderate driving, and aggressive driving, respectively; For the first i A cluster, The principal component feature vector of the driving event. For the first i The centroid vector of each cluster.

9. The method for recognizing aggressive driving behavior of heavy-duty vehicles based on multidimensional kinematic features according to claim 6, characterized in that, In S4, the Pearson correlation coefficient is introduced to calculate the linear correlation between the degree of aggressiveness of driving behavior and risk event indicators; When the linear correlation coefficient is significantly positive, it is determined that there is a significant positive correlation between the identified aggressive driving and the high-risk attribute, thus confirming the effectiveness of the identification method.

Images