Inter-vehicle correlation traffic monitoring method based on spatiotemporal analysis and trajectory matching

Abstract

The present invention belongs to the technical field of traffic control, and relates to a method for monitoring the correlation between vehicles based on spatiotemporal analysis and trajectory matching. The method first collects structured spatiotemporal data of vehicles, and defines spatiotemporal data trajectories and spatiotemporal data points; then detects vehicle trajectory features to obtain trajectory similarity scores; then detects vehicle stop points to obtain stop point similarity scores; then adds the trajectory similarity scores and the stop point similarity scores to obtain the correlation scores between vehicles. The method not only combines the existing deep learning technology for trajectory similarity analysis, but also introduces stop point detection and matching technology, comprehensively analyzes the correlation between vehicles from two dimensions of spatiotemporal data trajectories and spatiotemporal data points, more comprehensively and accurately captures and analyzes the potential correlation between vehicles, and improves the accuracy and practicality of the monitoring system.

Description

Technical Field

[0001] The invention belongs to the technical field of traffic control and relates to a method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching. Background Art

[0002] Accurately analyzing the correlation between vehicles is of great significance in modern traffic management and security monitoring. By analyzing the correlation between vehicles, it is possible to timely discover and deal with traffic congestion problems, and optimize traffic flow by analyzing the driving paths and parking behaviors of vehicles. At the same time, identifying and tracking suspicious vehicle behaviors can prevent and quickly respond to traffic accidents. In addition, by analyzing the driving and parking patterns of vehicles, data support can be provided to provide a basis for urban planning and traffic management. Therefore, a technical solution is needed that can comprehensively and accurately capture and analyze the potential correlation between vehicles.

[0003] When analyzing vehicle correlation, existing technologies mainly focus on the similarity analysis of the overall trajectory, ignoring the comparison between independent spatiotemporal data points. As a result, in some specific scenarios, it is impossible to accurately judge the short-term interaction and potential correlation between vehicles. Specifically, there are the following disadvantages:

[0004] (1) Trajectory matching technologies based on spatiotemporal data have their own applicable scenarios and limitations. The method that combines road network information assumes that the trajectory points are highly matched with the known road network, which has poor universality but is suitable for specific scenarios. Methods based on trajectory point matching, such as Euler distance and Manhattan distance, are intuitive but have limited effects when the trajectory length or acquisition frequency is inconsistent. Improved methods such as DTW, LCSS, ERP, EDR, etc., although they solve some problems, have high computational complexity and are sensitive to noise. Calculation methods based on trajectory shape, such as Hausdorff distance and Frechet distance, can capture the global shape characteristics of the trajectory, but there are challenges in time complexity and robustness. Although these methods can handle situations where sequence lengths are inconsistent, they are still limited in dealing with noise and capturing richer trajectory features. Computational complexity and scope of application have become the main bottlenecks.

[0005] (2) Deep learning methods mainly focus on analyzing the overall similarity of vehicle trajectories, that is, analyzing from the perspective of continuous trajectory sequences. Although this method can capture the overall trend and pattern of the trajectory, it often ignores the independent comparison and analysis between spatiotemporal data points, which may lead to the inability to accurately judge the short-term interaction and correlation between vehicles in some scenarios. For example, the behavior of a vehicle stopping briefly at a certain location may not significantly affect the similarity of the entire trajectory, but it is an important factor in judging the correlation of vehicles. Existing methods lack in-depth analysis of these independent spatiotemporal data points, which limits their application in complex traffic environments;

[0006] (3) There are some shortcomings in the application of stop point detection methods and user similarity measurement methods. First, the existing stop point detection methods are derived from user social network analysis. When identifying vehicle stop behavior, they lack optimization for traffic-specific environments and are easily affected by noise and dynamic changes. Second, the existing user similarity measurement methods are usually analyzed based on a single standard and cannot fully capture and measure the multi-dimensional correlation in complex traffic environments. These methods have high computational complexity and low efficiency when processing large-scale, complex spatiotemporal data, and are sensitive to noise and abnormal data. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a vehicle-to-vehicle correlation traffic monitoring method based on spatiotemporal analysis and trajectory matching. It not only combines the existing deep learning technology for trajectory similarity analysis, but also introduces stay point detection and matching technology. It comprehensively analyzes the correlation between vehicles from two dimensions: spatiotemporal data trajectory and spatiotemporal data points, so as to more comprehensively and accurately capture and analyze the potential correlation between vehicles and improve the accuracy and practicality of the monitoring system.

[0008] To achieve the above object, the present invention adopts the following technical solution:

[0009] A method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching, comprising the following steps:

[0010] S1. Collect vehicle structured spatiotemporal data and define spatiotemporal data trajectories and spatiotemporal data points;

[0011] S2, detect vehicle trajectory features, first extract trajectory features, then use the Seq2Seq autoencoder module to reconstruct the mobile behavior sequence, and generate a fixed-length deep representation of the trajectory, and finally perform similarity analysis to obtain the trajectory similarity score;

[0012] S3, detecting the vehicle's stop point, first detecting the vehicle's stop area, then calculating the specific stop point in the stop area, obtaining the specific stop position and time information of the vehicle in the specific area, and finally performing similarity analysis to obtain the stop point similarity score;

[0013] S4. Add the trajectory similarity score and the stop point similarity score to obtain the correlation score between vehicles.

[0014] As a further technical solution of the present invention, the spatiotemporal data points defined in step S1 are x= A triplet of (latitude, longitude, time) represents a latitude-longitude position and a timestamp. The spatiotemporal data trajectory is the sequence of triples. ,in It is spatiotemporal data points, is the number of spatiotemporal data points.

[0015] As a further technical solution of the present invention, the specific process of extracting trajectory features in step S2 is:

[0016] S211, sliding window generation: The width and offset of the sliding window are respectively and indicates and satisfies , each trajectory record is assigned to two consecutive windows, thus capturing trajectory changes;

[0017] S212, feature extraction: extracting behavior features from the trajectory records in each sliding window, processing the extracted behavior features through the sliding window method, and converting each trajectory into a feature sequence;

[0018] S213, Behavior feature calculation: In each sliding window, the change of behavior features is reflected by the attribute difference between two consecutive records. Suppose a window contains Records, recorded as , the attributes of each record include speed and turning rate, and the extracted behavioral features include time interval , Position Change , speed change and turning rate changes ,in , a window containing R records has R-1 behavior change features ;

[0019] S214, speed and turning rate calculation: for each track record in step S213 , whose coordinates are ,in represents the latitude of the trajectory point, represents the longitude of the trajectory point, Indicates time, the first record Speed and turning rate Set to 0 to calculate the speed for subsequent records and turning rate :

[0020] ,

[0021] ;

[0022] S215, feature set generation: After calculating the feature changes in each sliding window, a set of feature change rates, i.e., movement behavior, is obtained. ,in , , ,in .

[0023] As a further technical solution of the present invention, the Seq2Seq autoencoder module in step S2 includes an encoder RNN and a decoder RNN, and its working process is:

[0024] S221, encoder RNN reads the input movement behavior sequence , and update the hidden state in turn : , until the last input After that, hide the state is used as the representation of the entire sequence, and the hidden state is represented by the mobile action vector z ; The encoder takes the input movement behavior sequence Mapped to a fixed-length mobile behavior vector z, the mobile behavior vector z represents the spatiotemporal characteristics of the trajectory;

[0025] S222, decoder RNN generates output , and then As the initial hidden state, generate the subsequent output : ,in The decoder is The hidden state at step , is used to generate the One of the initial conditions for the step output, The decoder is The hidden state when generating the output, Depends on the previous hidden state And the current input, the RNN calculates the function Updated, is tanh or ReLU; the goal of the decoder is to reconstruct the input sequence , that is, the encoder RNN and decoder RNN are trained by minimizing the reconstruction error, minimizing the reconstruction error ; The decoder reconstructs the original input sequence from z. Since the input sequence itself is the learning target, the training process does not require any labeled data.

[0026] As a further technical solution of the present invention, the process of similarity analysis in step S2 is as follows: Suppose there are two trajectory feature vectors x and y, the new weighted distance D(x, y) is defined as:

[0027] ,

[0028] in, represents the Euclidean distance, represents the Manhattan distance, represents the cosine distance, , , are the weights of the three distance metrics, satisfying , Euclidean distance , Manhattan distance , cosine distance , ; The trajectory similarity score is obtained as: , the closer the score is to 1, the higher the similarity between the two trajectories.

[0029] As a further technical solution of the present invention, the vehicle stop area in step S3 is a geographical cluster area, which is a distance between the user and the vehicle. Stay time When detecting the vehicle stop area, the comprehensive index is calculated by combining the density-based method and the density-based method. The comprehensive index = w×density-based method+(1-w)×time-space constraint-based method, where w is a hyperparameter that can be adjusted according to actual conditions.

[0030] As a further technical solution of the present invention, the process of calculating the stop point in step S3 is: using the sliding window method, traversing the trajectory data, detecting the specific point where the vehicle stops in a certain area, and calculating the position of the stop point according to the average position of the trajectory points in each stop area. The obtained stop point position coordinates are: latitude ,longitude ; Time coordinate: arrival time , departure time .

[0031] As a further technical solution of the present invention, the similarity score of the stay point obtained in step S3 is ,in Two stopover points , The distance similarity of , Two stopover points , The time similarity of , is the KL divergence, , , are the weights of distance similarity and time similarity, respectively. .

[0032] As a further technical solution of the present invention, the correlation score between the vehicles in step S4 is , is a hyperparameter used to balance the impact of trajectory and stop points.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] (1) Comprehensive trajectory matching technology can capture and analyze the potential correlation between vehicles more comprehensively and accurately. The sliding window technology can not only capture the global trend of the trajectory, but also capture the details of local changes, thereby enhancing the all-round understanding of the trajectory. In addition, by combining multiple distance measurement methods such as Euclidean distance, Manhattan distance and cosine distance, while maintaining computational efficiency, it enhances the robustness to trajectories of different lengths and noise. By combining the trajectory shape-based measurement method with spatiotemporal feature extraction, it can more accurately measure the global shape and detailed changes of vehicle trajectories, thereby improving its applicability in complex traffic environments.

[0035] (2) The deep learning method adopted not only focuses on the overall similarity of trajectories, but also introduces the comparison and analysis of independent spatiotemporal data points. It can capture the vehicles’ stop behaviors and short-term interactions at key locations, and integrate the overall trajectory trends with the characteristics of key spatiotemporal points. The deep learning model captures complex spatiotemporal patterns, thereby improving the ability to recognize short-term correlations between vehicles. It can dynamically adjust the attention weights for the overall trajectory and local spatiotemporal points, further improving the flexibility and applicability of the model.

[0036] (3) An improved stay point detection and similarity measurement method is adopted to meet the specific needs of traffic monitoring. The density-based method and the spatiotemporal constraint-based method are combined, and the stay area is detected through a customized comprehensive indicator. This not only effectively processes the spatial characteristics of the trajectory, but also improves the detection accuracy through the dual constraints of time and space, thereby more accurately identifying the vehicle's stay area. In the calculation of stay point similarity, the Euclidean distance and Kullback-Leibler (KL) divergence are used to measure the spatial and temporal similarities respectively, and the final similarity score is obtained by weighted summation. This multi-dimensional and multi-level analysis method greatly improves the monitoring accuracy and practical application capabilities of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of spatiotemporal data trajectory and spatiotemporal data points defined in an embodiment of the present invention.

[0038] Figure 2 This is a flow chart of the vehicle-to-vehicle correlation traffic control method based on spatiotemporal analysis and trajectory matching of the present invention.

[0039] Figure 3 This is a flow chart of extracting trajectory features according to an embodiment of the present invention.

[0040] Figure 4This is a structural block diagram of the Seq2Seq autoencoder module described in the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described below by way of embodiments in conjunction with the accompanying drawings.

[0042] Example:

[0043] like Figure 2 As shown, this embodiment provides a method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching, which specifically includes the following steps:

[0044] S1. Vehicle structured spatiotemporal data collection:

[0045] like Figure 1 As shown, the space-time data trajectory and space-time data points are defined: the space-time data points are x= A triplet of (latitude, longitude, time) represents a latitude-longitude position and a timestamp. The spatiotemporal data trajectory is the sequence of triples. ,in It is spatiotemporal data points, is the number of spatiotemporal data points.

[0046] S2. Vehicle trajectory feature detection:

[0047] S21, trajectory feature extraction, uses a sliding window to traverse the trajectory record and extract features from each window to describe the movement behavior of the object, such as Figure 3 As shown in Figure 1, the sliding window method is used to obtain time-space invariant features to describe the mobile behavior. The specific process is as follows:

[0048] S211, sliding window generation: The width and offset of the sliding window are respectively and indicates and satisfies , each trajectory record is assigned to two consecutive windows, thus capturing trajectory changes;

[0049] S212, feature extraction: extracting behavior features from the trajectory records in each sliding window, processing the extracted behavior features through the sliding window method, and converting each trajectory into a feature sequence;

[0050] S213, Behavior feature calculation: In each sliding window, the change of behavior features is reflected by the attribute difference between two consecutive records. Suppose a window contains Records, recorded as , the attributes of each record include speed and turning rate, and the extracted behavioral features include time interval , Position Change , speed change and turning rate changes ,in , a window containing R records has R-1 behavior change features ;

[0051] S214, speed and turning rate calculation: Although the speed and turning rate are not explicitly listed in the original record, they can be calculated based on the position information. For each track record in step S213 , whose coordinates are ,in represents the latitude of the trajectory point, represents the longitude of the trajectory point, Indicates time, the first record Speed and turning rate Set to 0 to calculate the speed for subsequent records and turning rate :

[0052] ,

[0053] ;

[0054] S215, feature set generation: After calculating the feature changes in each sliding window, a set of feature change rates, i.e., movement behavior, is obtained. ,in , , ,in .

[0055] The S22 and Seq2Seq autoencoder modules reconstruct the movement behavior sequence and generate a fixed-length deep representation of the trajectory, such as Figure 4 As shown in Figure 1, the Seq2Seq autoencoder module includes an encoder RNN and a decoder RNN. The specific process is as follows:

[0056] S221, encoder RNN reads the input movement behavior sequence , and update the hidden state in turn : , until the last input After that, hide the state is used as the representation of the entire sequence, and the hidden state is represented by the mobile action vector z ; The encoder takes the input movement behavior sequence Mapped to a fixed-length mobile behavior vector z, the mobile behavior vector z represents the spatiotemporal characteristics of the trajectory;

[0057] S222, decoder RNN generates output , and then As the initial hidden state, generate the subsequent output : , The decoder is The hidden state at step , is used to generate the One of the initial conditions for the step output, The decoder is The hidden state when generating the output. It depends on the previous hidden state And the current input, the RNN calculates the function Updated, Can be tanh or ReLU; the goal of the decoder is to reconstruct the input sequence , that is, the encoder RNN and decoder RNN are trained by minimizing the reconstruction error, minimizing the reconstruction error ; The decoder reconstructs the original input sequence from z. Since the input sequence itself is the learning target, the training process does not require any labeled data.

[0058] S23, perform similarity analysis to obtain a trajectory similarity score; the specific process is: suppose there are two trajectory feature vectors x and y, and the new weighted distance D(x, y) is defined as:

[0059] ,

[0060] in, represents the Euclidean distance, represents the Manhattan distance, represents the cosine distance, , , are the weights of the three distance metrics, satisfying , Euclidean distance , Manhattan distance , cosine distance , , so the trajectory similarity score is: , the closer the score is to 1, the higher the similarity between the two trajectories.

[0061] S3. Vehicle stop point detection:

[0062] S31, stay area detection, stay area detection is to identify the time period of a vehicle staying in a certain geographical area. This embodiment is based on the density method and the method based on time and space constraints, and proposes a custom comprehensive indicator for detection. The stay area is a geographical cluster area, where the user is within the distance Stayed for a while ,This area can contain a series of locations, such as , which satisfies and , Indicates location and The distance between Indicates that the user is from a location Move to The density-based method in this embodiment adopts the density-based clustering algorithm DBSCAN, which uses the physical parameters of the trajectory, such as speed, acceleration and direction change, to detect the vehicle stop area through the sliding window technology. The core idea of ​​this method is that if there are enough points around a point within a certain distance, these points can be clustered into a cluster, indicating that the vehicle stays in this area; the method based on time-space constraints takes into account the dual constraints of time and space. If a sub-trajectory stays in a certain spatial area for more than a certain time threshold and within a certain distance threshold, then the area is regarded as a stop area. Specifically, for a trajectory , if satisfied , and , then the area is identified as a stay area; combining the two detection methods, a custom comprehensive index is proposed, and the calculation formula of the comprehensive index is as follows: Comprehensive index = w ×Density-based method+(1- w )×method based on spatiotemporal constraints, where w is a hyperparameter that can be adjusted according to actual conditions to optimize the effect of staying area detection. This comprehensive method not only takes advantage of the density-based method, which can effectively process the spatial characteristics of the trajectory, but also combines the method based on spatiotemporal constraints to improve the detection accuracy through dual constraints of time and space, thereby more accurately identifying the vehicle's staying area.

[0063] S32, stop point calculation, after identifying the vehicle's stop area, it is necessary to calculate the specific stop points in these areas. The stop point is the specific stop location and time information of the vehicle in a specific area. The stop point represents a geographical area where the vehicle stays for more than a given time threshold. , and at the distance threshold A stop point can be represented as a four-tuple , which includes the latitude of the stop point ,longitude and vehicle arrival time and departure time After identifying the stay area, take the average value of the position coordinates of all members in the stay area and take the earliest arrival time as , the latest departure time is Specifically, we use the sliding window method to traverse the trajectory data and detect the specific points where the vehicle stops in a certain area. According to the average position of the trajectory points in each stop area, we calculate the position of the stop point. The coordinates of the stop point are: latitude ,longitude ; Time coordinate: arrival time , departure time , thereby calculating the specific stop points in each stop area and providing detailed stop information of vehicles in a specific area.

[0064] S33, similarity analysis, the stop point similarity calculation module is used to evaluate the similarity between different vehicle stop points. The stop point similarity score consists of two parts: distance similarity and time similarity. Use Euclidean distance to measure two stop points , The space distance between: The time similarity uses Kullback-Leibler (KL) divergence to measure the difference in time distribution between two stay points. Assume that the two stay points , The time distribution is and , then the temporal similarity calculation formula is: , where KL divergence Defined as: , the final distance similarity score combines distance similarity and time similarity and is calculated by weighted summation: ,in, , are the weights of distance similarity and time similarity, respectively. .

[0065] S4. Add the trajectory similarity score and the stop point similarity score to obtain the correlation score between vehicles. Specifically, the correlation score between vehicles is , is a hyperparameter used to balance the influence of trajectory and stop points.

[0066] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0067] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching, characterized in that: The specific steps include: S1. Collect vehicle structured spatiotemporal data and define spatiotemporal data trajectories and spatiotemporal data points; S2. Detect vehicle trajectory features. First, extract trajectory features, then use the Seq2Seq autoencoder module to reconstruct the mobile behavior sequence, and generate a fixed-length depth representation of the trajectory. Finally, perform similarity analysis to obtain the trajectory similarity score. The process of similarity analysis is as follows: Suppose there are two trajectory feature vectors x and y, and the new weighted distance D(x, y) is defined as: ， in, represents the Euclidean distance, represents the Manhattan distance, represents the cosine distance, , , are the weights of the three distance metrics, satisfying , Euclidean distance , Manhattan distance , cosine distance , ; The trajectory similarity score is obtained as: ,The closer the score is to 1, the higher the similarity of the two trajectories; S3, detecting the vehicle's stop point, first detecting the vehicle's stop area, then calculating the specific stop point in the stop area, obtaining the specific stop position and time information of the vehicle in the specific area, and finally performing similarity analysis to obtain the stop point similarity score; the stop point similarity score is ,in Two stopover points , The distance similarity of , , Stop points , The latitude, , Stop points , longitude; Two stopover points , The time similarity of , is the KL divergence, , , are the weights of distance similarity and time similarity, respectively. ; S4. Add the trajectory similarity score and the stop point similarity score to obtain the correlation score between vehicles.

2. The method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching according to claim 1, characterized in that: The spatiotemporal data points defined in step S1 are x= A triplet of (latitude, longitude, time) represents a latitude-longitude position and a timestamp. The spatiotemporal data trajectory is the sequence of triples. ,in It is spatiotemporal data points, is the number of spatiotemporal data points.

3. The method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching according to claim 2 is characterized in that: The specific process of extracting trajectory features in step S2 is as follows: S211, sliding window generation: The width and offset of the sliding window are respectively and indicates and satisfies , each trajectory record is assigned to two consecutive windows, thus capturing trajectory changes; S212, feature extraction: extracting behavior features from the trajectory records in each sliding window, processing the extracted behavior features through the sliding window method, and converting each trajectory into a feature sequence; S213, Behavior feature calculation: In each sliding window, the change of behavior features is reflected by the attribute difference between two consecutive records. Suppose a window contains Records, recorded as , the attributes of each record include speed and turning rate, and the extracted behavioral features include time interval , Position Change , speed change and turning rate changes ,in , a window containing R records has R−1 behavior change features ; S214, speed and turning rate calculation: for each track record in step S213 , whose coordinates are ,in represents the latitude of the trajectory point, represents the longitude of the trajectory point, Indicates time, the first record Speed and turning rate Set to 0 to calculate the speed for subsequent records and turning rate : ， ； S215, feature set generation: After calculating the feature changes in each sliding window, a set of feature change rates, i.e., movement behavior, is obtained. ,in , , ,in .

4. The method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching according to claim 3 is characterized in that: The Seq2Seq autoencoder module in step S2 includes an encoder RNN and a decoder RNN, and its working process is as follows: S221, encoder RNN reads the input movement behavior sequence , and update the hidden state in turn : , until the last input After that, hide the state is used as the representation of the entire sequence, and the hidden state is represented by the mobile action vector z ; The encoder takes the input movement behavior sequence Mapped to a fixed-length mobile behavior vector z, the mobile behavior vector z represents the spatiotemporal characteristics of the trajectory; S222, decoder RNN generates output , and then As the initial hidden state, generate the subsequent output : ,in The decoder is The hidden state at step , is used to generate the One of the initial conditions for the step output, The decoder is The hidden state when generating the output, Depends on the previous hidden state And the current input, the RNN calculates the function Updated, is tanh or ReLU; the goal of the decoder is to reconstruct the input sequence , that is, the encoder RNN and decoder RNN are trained by minimizing the reconstruction error, minimizing the reconstruction error .

5. The method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching according to claim 4 is characterized in that: The vehicle stop area in step S3 is a geographical cluster area, which is the distance between the user and the vehicle. Stay time When detecting the vehicle stop area, the comprehensive index is calculated by combining the density-based method and the density-based method. The comprehensive index = w×density-based method+(1−w)×time-space constraint-based method, where w is a hyperparameter that can be adjusted according to actual conditions.

6. The method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching according to claim 5, characterized in that: The process of calculating the stop point in step S3 is: using the sliding window method, traversing the trajectory data, detecting the specific point where the vehicle stops in a certain area, and calculating the position of the stop point based on the average position of the trajectory points in each stop area. The obtained stop point position coordinates are: latitude ,longitude ; Time coordinate: arrival time , departure time .

7. The method for monitoring inter-vehicle correlation traffic based on spatiotemporal analysis and trajectory matching according to claim 6, characterized in that: Step S4: The correlation score between vehicles , is a hyperparameter used to balance the impact of trajectory and stop points.

Images