A highway section safety risk state cognition method based on trajectory data

Abstract

The application provides a highway section safety risk state cognition method based on trajectory data. The method first selects a highway section as a research area, obtains the vehicle types and trajectory data of all vehicles passing through the research area within a certain period of time, distributes the motion parameters of the vehicles to each grid after gridding the research area, performs conflict analysis based on the trajectory data and alternative safety measures and obtains a conflict index TIT, divides the risk labels of the TIT distribution of all samples by using a quantile method, and finally constructs a time-space sequence deep learning model taking grid map data as input and risk labels as output to identify the risk state and perform influence analysis on the time and space levels. The application is helpful for real-time identification of potential traffic safety risks on the road and provides support for realizing a safe, efficient and intelligent highway traffic system.

Description

Technical Field

[0001] This invention relates to a method for recognizing road segment safety risk status based on trajectory data, belonging to the field of traffic safety research technology. Background Technology

[0002] The purpose of road traffic safety risk assessment is to identify potential risks on the road and provide relevant preventative measures to reduce the occurrence and severity of traffic accidents. By assessing the traffic safety risk status, traffic management departments can understand the safety conditions on the road and take measures to improve traffic safety, including improving road design and traffic facilities, strengthening the enforcement of traffic regulations, raising drivers' traffic safety awareness, and providing emergency rescue and response.

[0003] Currently, the analysis of road segment safety risk status mainly relies on macro-level traffic flow data (flow rate, speed, occupancy, etc.) for statistical analysis or on artificial intelligence technology for risk identification and prediction. However, this traffic flow data often overlooks the micro-level behaviors of many vehicles and the interactions between them. Therefore, analyzing trajectory data can capture more micro-level vehicle behaviors, which helps to fully analyze traffic safety levels. Meanwhile, spatiotemporal sequence data is complex high-dimensional data. Leveraging the powerful spatial and temporal feature learning capabilities of spatiotemporal sequence deep learning models, the extracted features can more effectively characterize traffic safety risk levels, showing great potential for real-time identification of traffic safety risks. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a method, system, device, and storage medium for recognizing the safety risk status of highway sections based on trajectory data.

[0005] In a first aspect, the present invention provides a method for recognizing the safety risk status of highway sections based on vehicle trajectory data, the method comprising the following steps:

[0006] c1. Select a section of highway as the study area and obtain the vehicle type and trajectory data of all vehicles passing through the study area within a certain period of time.

[0007] c2. Divide the road segment into a grid and assign the motion parameters of each vehicle at each moment to the grid according to the location information;

[0008] c3. Conduct conflict analysis based on trajectory data and alternative safety measures to obtain conflict index TIT. Use the quantile method to label the risk of conflict index TIT distribution for all samples.

[0009] c4. Construct a spatiotemporal sequence deep learning model that takes grid map data as input and risk labels as output to identify risk states and perform impact analysis at the temporal and spatial levels.

[0010] Secondly, the present invention provides a highway section safety risk status cognition system based on vehicle trajectory data, comprising:

[0011] The data acquisition module is used to select a section of highway as the study area and acquire vehicle type and trajectory data of all vehicles passing through the study area within a certain period of time.

[0012] The gridding module is used to divide the road segment into grids and allocate the motion parameters of each vehicle at each moment to the grid according to the location information.

[0013] The risk labeling module is used to perform conflict analysis based on trajectory data and alternative safety measures, obtain conflict index TIT, and use the quantile method to label the risk of conflict index TIT distribution for all samples.

[0014] The risk identification module is used to build a spatiotemporal sequence deep learning model that takes grid data as input and risk labels as output to identify risk states and perform impact analysis at both the temporal and spatial levels.

[0015] Thirdly, the present invention provides a highway section safety risk status recognition device based on vehicle trajectory data, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned highway section safety risk status recognition method based on vehicle trajectory data.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing a computer program for executing the above-described method for recognizing the safety risk status of highway sections based on vehicle trajectory data.

[0017] The beneficial effects of this invention are:

[0018] First, this invention utilizes micro-trajectory data to assess road segment safety risk status, which can capture more information about vehicle micro-behavior and the interaction between vehicles.

[0019] Secondly, this invention uses the quantile method to label the risk of all samples' TIT distribution. The labeling method is easy to calculate and does not require subjective intervention. Furthermore, the TTC threshold can be adjusted, which has the advantages of being objective and highly flexible.

[0020] Third, the CNN-Attention model used in this invention can capture features in both time and space simultaneously, and perform influence analysis at both time and space levels, which has the advantages of high accuracy and strong interpretability. Attached Figure Description

[0021] Figure 1 This is an overall framework diagram of the method of the present invention;

[0022] Figure 2 A schematic diagram illustrating road segment division and the allocation of a certain parameter at a specific moment;

[0023] Figure 3 This is a schematic diagram of the CNN-Attention model. Detailed Implementation

[0024] The main concept of this application is as follows: A section of highway is selected as the study area. Vehicle type and trajectory data of all vehicles passing through this study area within a certain time period are obtained. After gridding the study area, vehicle motion parameters are assigned to each grid cell. Conflict analysis is performed based on trajectory data and Surrogate Safety Measures (SSMs) to obtain the Time Integrated Traffic Response Index (TIT). The TIT distribution of all samples is used to classify risk labels using the quantile method. Finally, a spatiotemporal deep learning model is constructed, using grid data as input and risk labels as output, to identify risk states and perform impact analysis at both temporal and spatial levels. This method helps to identify potential traffic safety risks on roads in real time, providing support for achieving a safe, efficient, and intelligent highway traffic system.

[0025] This application provides a method for recognizing the safety risk status of highway sections based on vehicle trajectory data, including the following steps:

[0026] c1. Select a section of highway as the study area and obtain the vehicle type and trajectory data of all vehicles passing through the study area within a certain period of time.

[0027] c2. Divide the road segment into a grid and assign the motion parameters of each vehicle at each moment to the grid according to the location information;

[0028] c3. Conduct conflict analysis based on trajectory data and alternative safety measures to obtain the conflict index TIT, and use the quantile method to label the risk of the TIT distribution of all samples;

[0029] c4. Construct a spatiotemporal sequence deep learning model that takes grid map data as input and risk labels as output to identify risk states and perform impact analysis at the temporal and spatial levels.

[0030] In one example, in step c1:

[0031] After selecting a study road segment, vehicle type and trajectory data for all vehicles passing through the study area within a certain time period are obtained. Vehicle types include cars and trucks, and trajectory data includes vehicle relative coordinates, lateral velocity, longitudinal velocity, lateral acceleration, and longitudinal acceleration. Vehicle type data is categorical data, and trajectory data is time-series data collected at a frequency of 25 Hz. The road segment is then gridded into N... L ×N G The grid map (each grid is the width of a lane and 4.6m long).

[0032] In one example, in step c2:

[0033] The road segment is divided into grids. Based on the average length of cars and trucks, the length of a single grid is set to 4.6m, and the width is equal to the lane width. The final grid map size is as follows:

[0034] (N L ×w L )×(N G ×4.6)

[0035] Where, N L ×w L N represents the overall width of the grid. L w is the number of lanes L N represents the width of a single lane. G ×4.6 represents the total grid length, N G Let N be the number of grid cells for each lane. Then the shape of the grid is N. L ×N G .

[0036] The motion parameters of each vehicle at each moment are assigned to a grid. These parameters include lateral velocity, longitudinal velocity, lateral acceleration, and longitudinal acceleration. The assignment method is as follows: the vehicle motion parameters are mapped to each grid based on the vehicle's center point, with smaller vehicles occupying one grid, larger vehicles occupying two grids, and grids with no vehicles having a value of 0. A schematic diagram of a single grid is shown below. Figure 2 As shown.

[0037] In one example, in step c3:

[0038] c31. Calculate the alternative safety measure index TTC (Time To Collision).

[0039]

[0040] Where, x n-1 (t), x n (t) represents the front and rear positions of the vehicles, respectively, v n-1 (t), v n(t) represents the speeds of the front and rear vehicles, respectively, and L represents the length of the front vehicle.

[0041] c32. Calculate the Time Integrated TTC (TIT) conflict index.

[0042]

[0043] Among them, TTC n (t) represents the TTC value of the vehicle at time t, where TTC * The threshold for TTC is 1.5 to 4 seconds.

[0044] c33. Risk labeling based on quantile method

[0045] By summing the TIT values ​​of all samples, we can obtain the TIT distribution. The risk is then labeled using the quantile method: samples with a TIT less than the 15th quantile are considered safe; samples with a TIT between the 15th and 85th quantiles are considered relatively risky; and samples with a TIT greater than the 85th quantile are considered dangerous, as shown in the following formula:

[0046]

[0047] Where i is a certain sample, TIT 15％ TIT 85％ These are the 15th and 85th percentiles of the TIT distribution for all samples, with 0, 1, and 2 representing safe, relatively dangerous, and dangerous, respectively.

[0048] After obtaining all the data and corresponding labels, the dataset is divided into training set: test set ratio of 4:1.

[0049] In one example, in step c4:

[0050] c41. Model Building and Training

[0051] A spatiotemporal sequence deep learning model is constructed with grid map data as input and risk labels as output, and the model is trained using training set data.

[0052] Specifically, the input grid data is in the following format:

[0053]

[0054] Where t is the time step, m is the number of grids (i.e. the number of features), and NL and NG are the width and length of the grid, respectively.

[0055] The model uses a CNN-Attention approach. Data X first passes through a convolutional neural network (CNN), which contains only convolutional layers. The operation of the convolutional layers is as follows:

[0056] X CNN =ReLU(W*X) input +b)

[0057] Among them, X input X CNN The input and output data are respectively, W is the weight of the convolution kernel, * indicates the convolution operation, b is the bias term, and ReLU is the activation function.

[0058] Then X CNN The input will first undergo three linear transformations after passing through the Attention layer:

[0059] Q = W q ·X CNN +b q

[0060] K = W k ·X CNN +b k

[0061] V = W v ·X CNN +b v

[0062] Among them, W q W k W v and b q b k b v These are the weight term and the bias term, respectively.

[0063]

[0064] Among them, X A d is the output of the Attention layer. k Let k be the dimension. For attention weights, Softmax guarantees that the sum of the attention weights is 1.

[0065] Finally, the output of the attention layer is mapped to the output dimension through a fully connected layer, and then passed through the softmax function to output the corresponding class probability, as shown below:

[0066] X output =softmax(W fc ·X A +b fc )

[0067] Among them, W fc and b fc These are the weights and biases of the fully connected layer, respectively.

[0068] Due to class imbalance among samples, the model uses a weighted cross-entropy loss function, as shown below:

[0069] f loss =-(w1·y1·log(p1)+w2·y2·log(p2)+w3·y3·log(p3))

[0070] Where w1, w2, and w3 are the class weights, y1, y2, and y3 are the one-hot encodings of the true labels, and p1, p2, and p3 are the probability values ​​predicted by the model. The model structure diagram is attached. Figure 3 As shown.

[0071] c42. Model Testing and Impact Analysis

[0072] After obtaining the trained risk identification model, it is validated using samples from the test set, and the weights of the test set samples in time step and space are output for impact analysis.

[0073] The weights of the time steps are attention weights, i.e. The analysis is performed using Class Activation Mapping (CAM), as shown in the following equation:

[0074]

[0075] Among them, X CNN and These represent the output and attention weights of the CNN layer, respectively, with the subscript i indicating the i-th channel. The spatial weights are calculated by summing the weighted sum of the CNN layer outputs and their corresponding attention weights for all channels.

[0076] This application also discloses a highway section safety risk status recognition system based on vehicle trajectory data, including:

[0077] The data acquisition module is used to select a section of highway as the study area and acquire vehicle type and trajectory data of all vehicles passing through the study area within a certain period of time.

[0078] The gridding module is used to divide the road segment into grids and allocate the motion parameters of each vehicle at each moment to the grid according to the location information.

[0079] The risk labeling module is used to perform conflict analysis based on trajectory data and alternative safety measures, obtain the conflict index TIT, and use the quantile method to label the risk of the conflict index TIT distribution of all samples.

[0080] The risk identification module is used to build a spatiotemporal sequence deep learning model that takes grid data as input and risk labels as output to identify risk states and perform impact analysis at both the temporal and spatial levels.

[0081] This application also discloses a highway section safety risk status recognition device based on vehicle trajectory data, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned highway section safety risk status recognition method based on vehicle trajectory data.

[0082] This application also discloses a computer-readable storage medium storing a computer program for executing the above-described method for recognizing the safety risk status of highway sections based on vehicle trajectory data.

[0083] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory.

[0084] The non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. The volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0085] The processor in this application may include one or more processing cores. The processor executes instructions, programs, code sets, or instruction sets stored in memory, and calls data stored in memory to perform various functions and process data. The processor may be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), controller, microcontroller, and microprocessor. It is understood that, for different devices, the electronic devices used to implement the above-described processor functions may also be other types, and the embodiments of this application do not specifically limit this.

[0086] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0087] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for recognizing the safety risk status of highway sections based on vehicle trajectory data, characterized in that, The method includes the following steps: c1. Select a section of highway as the study area and obtain the vehicle type and trajectory data of all vehicles passing through the study area within a certain period of time. c2. Divide the road segment into a grid and assign the motion parameters of each vehicle at each moment to the grid according to the location information; c3. Conduct conflict analysis based on trajectory data and alternative safety measures to obtain conflict index TIT. Use the quantile method to label the risk of conflict index TIT distribution for all samples. c4. Construct a spatiotemporal sequence deep learning model with grid map data as input and risk labels as output to identify risk status and conduct impact analysis at the temporal and spatial levels; Step c3 specifically includes: c31. Calculate the alternative safety measure index (TTC). in, These are the front and rear positions of the vehicles, respectively. These represent the speeds of the vehicles in front and behind, respectively, and L is the length of the vehicle in front. c32. Calculate the conflict index TIT in, Let be the TTC value of the vehicle at time t. The threshold for TTC is 1.5 to 4 seconds. c33. Risk labeling based on quantile method The conflict index TIT of all samples is summarized to obtain the distribution of conflict index TIT. The risk is marked by the quantile method. Conflict index TIT is less than the 15th quantile. The sample is safe. The sample is more dangerous when the conflict index TIT is between the 15th and 85th quantile. The sample is dangerous when the conflict index TIT is greater than the 85th quantile. After obtaining all the data and corresponding labels, the dataset is divided into training set: test set = 4:

1.

2. The method for recognizing road segment safety risk status according to claim 1, characterized in that, The vehicle types obtained in step c1 include small cars and large vehicles, and the trajectory data includes the vehicle's relative coordinates, lateral velocity, longitudinal velocity, lateral acceleration, and longitudinal acceleration.

3. The method for recognizing road segment safety risk status according to claim 2, characterized in that, The vehicle type is category data, and the trajectory data is time series data collected at a frequency of 25 Hz.

4. The method for recognizing road segment safety risk status according to claim 2, characterized in that, Step c2 involves gridding the road segment. Specifically, based on the average length of cars and trucks, the length of a single grid is set to 4.6m, and the width is equal to the lane width. The final grid map size is as follows: in, The overall width of the grid. For the number of lanes, The width of a single lane; The total length of the grid. Let be the number of grid cells for each lane; then the shape of the grid is: .

5. The method for recognizing road segment safety risk status according to claim 4, characterized in that, In step c2, the motion parameters of each vehicle at each moment are assigned to a grid. The motion parameters include lateral velocity, longitudinal velocity, lateral acceleration, and longitudinal acceleration. Specifically, the assignment method is as follows: the vehicle motion parameters are mapped to each grid according to the vehicle's center point, where small vehicles occupy one grid, large vehicles occupy two grids, and the grid data for no vehicle is 0.

6. The method for recognizing road segment safety risk status according to claim 1, characterized in that, Step c4 specifically includes: c41. Model Building and Training A spatiotemporal sequence deep learning model is constructed with grid map data as input and risk labels as output, and the model is trained using training set data; Specifically, the input grid data is in the following format: Where t is the time step and m is the number of grid plots. and These are the width and length of the grid, respectively; The model uses a CNN-Attention approach, where data X is processed through a convolutional neural network (CNN) to become... ; After passing through the Attention layer, we obtain ;Will The output dimension is mapped through a fully connected layer, and then the corresponding class probability is output through a softmax function. c42. Model Testing and Impact Analysis The test set samples are used for validation, and the weights of the test set samples in time step and space are output for impact analysis. The weights in time step are analyzed using attention weight row analysis, and the weights in space are calculated using class activation mapping, that is, by calculating the weighted sum of the outputs of all channels of CNN layers and the corresponding attention weights.

7. A highway section safety risk status recognition system based on vehicle trajectory data, used to implement the method of claim 1, characterized in that, include: The data acquisition module is used to select a section of highway as the study area and acquire vehicle type and trajectory data of all vehicles passing through the study area within a certain period of time. The gridding module is used to divide the road segment into grids and allocate the motion parameters of each vehicle at each moment to the grid according to the location information. The risk labeling module is used to perform conflict analysis based on trajectory data and alternative safety measures, obtain conflict index TIT, and use the quantile method to label the risk of conflict index TIT distribution for all samples. The risk identification module is used to build a spatiotemporal sequence deep learning model that takes grid data as input and risk labels as output to identify risk states and perform impact analysis at both the temporal and spatial levels.

8. A highway section safety risk status recognition device based on vehicle trajectory data, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method for recognizing the safety risk status of a highway section based on vehicle trajectory data as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which is used to execute the highway section safety risk status recognition method based on vehicle trajectory data as described in any one of claims 1-6.

Images