A method and system for curve collision warning of connected and cooperative driving vehicles

Abstract

This invention belongs to the field of autonomous driving technology, specifically relating to a method and system for curve collision warning of connected and cooperative driving vehicles. The method includes: collecting vehicle status information and curve environment information from roadside RSUs and uploading them to an edge cloud; the edge cloud performs fusion analysis on the input data and uses a Bi-LSTM network to predict the driving trajectory of non-connected vehicles at the curve; establishing a driving risk field model for intelligent connected vehicles, and combining the predicted trajectories of surrounding vehicles to assess the collision risk of curve driving in real time; using the Monte Carlo method and combining human driving data to calculate the collision risk threshold for curve collision warning; this invention, by enabling intelligent L2-level intelligent vehicles to perceive the environment through connectivity and establishing a curve collision warning model, can effectively reduce the problems of missed and false alarms in collision warning systems, thereby improving driving safety and efficiency.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, specifically relating to a method and system for warning of curve collisions in connected and cooperative driving vehicles. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, the level of intelligence in automobiles is constantly improving. More and more vehicles are interacting with their surroundings through V2X (Vehicle-to-Everything) technology, greatly improving road traffic safety and driving efficiency. However, the complexity of curves limits the perception range of traditional onboard sensors (such as radar and cameras) and the influence of the external environment, resulting in blind spots and making collision warning systems less effective in such scenarios.

[0003] While existing cornering collision warning systems offer a degree of safety, they typically rely on the vehicle's own sensing equipment and cannot acquire real-time information about other vehicles, limiting the accuracy and reliability of their warnings. In contrast, vehicle-to-everything (V2X) systems based on intelligent connected vehicle technology can overcome these limitations through real-time information sharing, providing more comprehensive environmental perception and more accurate risk assessment.

[0004] In recent years, with the introduction of artificial intelligence technology, especially the application of advanced algorithms such as Bidirectional Long Short-Term Memory (Bi-LSTM) networks, connected and cooperative driving technology has made significant progress in risk prediction and assessment in complex scenarios. These technologies can accurately predict the driving trajectory of non-connected vehicles at curves and, combined with the driving risk field model of intelligent connected vehicles, significantly improve driving safety and efficiency. However, existing technologies still have certain limitations. Specifically, the safety and efficiency of environmental perception driving in current connected and L2-level intelligent vehicles are not high enough, and collision warnings at curves are not reliable enough, with issues of missed and false alarms. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a method and system for warning of cornering collisions in connected and cooperative driving vehicles.

[0006] On one hand, the present invention provides a method for curve collision warning of connected cooperative driving vehicles, comprising the following steps:

[0007] S1. Collect roadside background vehicle status information and curve environment information, and upload them to EC;

[0008] S2 and EC fuse the input data and use a Bi-LSTM network to predict the driving trajectory of non-connected vehicles at curves;

[0009] S3. Establish a driving risk field model for intelligent connected vehicles, and combine it with the predicted trajectories of surrounding vehicles to assess the collision risk of driving on curves in real time.

[0010] S4. Utilize the Monte Carlo method and combine it with human driving data to calculate the collision risk threshold and provide cornering collision warning.

[0011] On the other hand, the present invention provides a connected and cooperative driving vehicle curve collision warning system, which includes: a status information acquisition unit, a trajectory prediction unit, a risk assessment module, and a collision warning module.

[0012] The status information acquisition unit is used to acquire vehicle status information, surrounding vehicle status information, and road environment information;

[0013] The trajectory prediction unit predicts the driving trajectories of non-connected vehicles around the vehicle based on the input information from the information acquisition unit.

[0014] The risk assessment module calculates the risk of the vehicle's cornering collision in real time based on the vehicle's information, surrounding vehicle information, and road environment information.

[0015] The collision avoidance warning module compares the real-time collision risk of the vehicle on the curve with a built-in collision risk threshold. When the risk exceeds the threshold, a collision risk warning is issued.

[0016] The beneficial effects of this invention are:

[0017] This invention enhances the environmental perception of Level 2 intelligent vehicles through network connectivity, establishing a collision warning model for cornering, thereby improving driving safety and efficiency. By fusing multi-source data and utilizing a Bi-LSTM network model, it accurately predicts the driving trajectory of non-networked vehicles at corners, and combines this with a driving risk field model for intelligent connected vehicles to assess collision risks in real time. By using the Monte Carlo method combined with human driving data to calculate collision risk thresholds, the collision warning becomes more reliable, effectively reducing false alarms and missed alarms in collision warning systems. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for curve collision warning of connected and cooperative driving vehicles according to the present invention;

[0019] Figure 2 This is a schematic diagram of the curve collision warning system for connected and cooperative driving vehicles according to the present invention.

[0020] Figure 3 This is a schematic diagram of the system structure for the curve collision warning of a connected and cooperative driving vehicle according to the present invention.

[0021] Figure 4This is a system technical roadmap for the curve collision warning of connected and cooperative driving vehicles according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, the present invention provides a method for curve collision warning of a connected cooperative driving vehicle, comprising the following steps:

[0024] S1. The roadside RSU collects vehicle status information and curve environment information and uploads it to the edge cloud (EC);

[0025] like Figure 1 In step S1, the roadside unit (RSU) is installed at the curve entrance or other key locations. Utilizing its multi-sensor system, such as high-precision GPS, LiDAR, millimeter-wave radar, and high-definition cameras, it continuously monitors vehicles passing through the curve and their surrounding environment. Vehicle status information includes, but is not limited to, vehicle position, speed, steering wheel angle, heading angle, acceleration, vehicle dimensions, and its lateral position relative to the lane. Curve environment information covers lane curvature, road width, road surface conditions, real-time weather information (such as road slipperiness due to rain or snow), and lane marking information (such as lane lines and signs). By integrating multi-sensor data, the RSU can generate comprehensive environmental perception data and transmit this data to the EC (Engineer) for low-latency processing via a 5G communication network.

[0026] S2 and EC fuse and analyze the input data, and use a bidirectional long short-term memory network (Bi-LSTM) model to predict the driving trajectory of non-connected vehicles at curves;

[0027] like Figure 1 In the S2 and EC modules, when fusing and parsing the data uploaded by the RSU, a unified dataset is formed through temporal alignment, spatial mapping, and feature extraction. Then, a Bi-LSTM model is used to predict the driving trajectories of non-connected vehicles at curves. The Bi-LSTM model captures more temporal features by independently processing forward and backward time-series data, thus improving prediction accuracy. The model fuses forward and backward feature vectors to form a more reliable trajectory prediction.

[0028] After data fusion, the edge cloud uses a Bi-LSTM model to process the fused data. The Bi-LSTM model consists of two independent LSTM networks: one for processing forward time series data and the other for processing backward time series data. This bidirectional processing captures more temporal features and enhances the accuracy of trajectory prediction.

[0029] Specifically, the processing steps for the Bi-LSTM model are as follows:

[0030] Step 1: Process the sequence from x1 to x using a forward LSTM. t And generate a hidden state at each time step t.

[0031] Step 2: Process the sequence from x using inverse LSTM. t To x1, and generate a hidden state at each time step t.

[0032] Step 3: For each time step t, combine the feature vectors extracted by the forward and backward LSTMs. and The parts are concatenated together to form a single feature vector h. t ;

[0033] Using a weighted sum to merge feature vectors can be represented as:

[0034]

[0035] Here, α is a weighting parameter (usually between 0 and 1) used to control the contribution of positive and negative feature vectors during merging.

[0036] These feature vectors are then input into a hidden layer for deeper feature extraction; the expression is:

[0037] f = W·h t +c

[0038] a = σ(f)

[0039] Where W is the weight matrix; c is the bias term; and σ is the activation function Sigmoid.

[0040] S3. Establish a driving risk field model for intelligent connected vehicles, and combine it with the predicted trajectories of surrounding vehicles to assess the collision risk of driving on curves in real time.

[0041] like Figure 1In section S3, based on the non-connected vehicle trajectory predicted by the aforementioned Bi-LSTM model, the system further establishes a driving risk field model for intelligent connected vehicles. This model uses a Gaussian function to quantify the collision risk of the vehicle at different locations. Model parameters such as height and width are dynamically adjusted by factors like vehicle speed and road curvature to accurately reflect the current driving risk. By combining the predicted trajectory of the non-connected vehicle with environmental information, the system can dynamically adjust the parameters of the Gaussian function to generate accurate risk field maps. These maps combine the vehicle's information with the trajectories of surrounding vehicles, providing a foundation for real-time collision risk assessment.

[0042] Specifically, the construction of the driving risk field model includes the following steps:

[0043] Step 1: Determine the parameters of the risk field model. The risk field model uses a two-dimensional Gaussian function to represent the collision risk at different locations. The model's risk function R(x,y) represents the collision risk at the coordinate point (x,y), and is defined as:

[0044]

[0045] Where R(x,y) is the collision risk function for point (x,y), h represents the height of the Gaussian cross section function, w represents the width of the Gaussian cross section function, and r c Represents the vehicle's turning radius, (x o ,y o ) represents the coordinates of the center point of the turning radius.

[0046] Step two involves combining vehicle trajectory with environmental information. The risk field model integrates the predicted trajectory of the non-connected vehicle, the current vehicle's dynamic information (such as speed and direction), and curve environment information (such as road curvature and width). This information is used to dynamically adjust the parameters of the Gaussian function, ensuring that the risk field accurately reflects the current driving risk. To better achieve this, the system first generates an initial map based on the road environment and obstacles, using the surrounding environmental information of the intelligent connected vehicle's driving scenario, and then divides this map into multiple grid points.

[0047] Next, the grid points are assigned values ​​and their costs are calculated based on the obstacle type and orientation. Specifically, the system defines costs according to the obstacle type and combines these costs with the obstacle's orientation to generate a driving scenario cost map. The orientation cost is calculated based on the coordinates of all grid points, vehicle position, road width, and vehicle turning radius. Updating the orientation matrix using the Dirac delta function further optimizes and reflects the orientation cost of obstacles. The updated orientation matrix is ​​expressed as:

[0048]

[0049] Among them, Hdir,i This is a position matrix representing the position information of surrounding objects relative to the vehicle. `dir` represents different orientations of objects relative to the vehicle. i The initial orientation matrix is ​​calculated by combining the coordinates of all points on the current cost map with the vehicle position, road width, and vehicle turning radius. This initial orientation matrix represents the sensor's sensing range, which is determined by dividing the map into grid points. The smaller the sensing range, the greater the computational load of the grid points involved.

[0050] The driving scenario cost is obtained by multiplying the cost of each obstacle type and the cost of the obstacle's location. The driving scenario cost Sc is denoted as...

[0051]

[0052] Where Sc represents the driving scenario cost of the intelligent connected vehicle, dir=1 represents the front, dir=2 represents the rear, dir=3 represents the left, dir=4 represents the right, i represents the obstacle type, and cost i The value represents the cost of defining the i-th type of obstacle; i=1 represents edge vegetation, i=2 represents roads, and i=3 represents obstacle vehicles.

[0053] Multiply the risk field by the cost graph of the driving scenario, and sum over all grid points to obtain the quantified driving risk R. k :

[0054]

[0055] Among them, R k This represents the driving collision risk value, and meshgrids represents the set of all grid points.

[0056] Step 3: Real-time Risk Assessment. By calculating the risk value R(x,y) at different locations, the system can assess the collision risk at the vehicle's location in real time during its movement. This risk assessment considers not only the current position and speed of the vehicle but also the trajectories of surrounding vehicles, thus providing a comprehensive risk assessment result.

[0057] S4. Utilize the Monte Carlo method and combine it with human driving data to calculate the collision risk threshold and provide cornering collision warning.

[0058] like Figure 1In the S4 system, the Monte Carlo method is used to simulate various driving scenarios that may be encountered during cornering, in order to assess the collision risk under different scenarios. The Monte Carlo method comprehensively captures various risk events that a vehicle may encounter while cornering by generating a large number of random sample driving scenarios. These scenarios include parameters such as the relative position and speed of the vehicle with respect to surrounding obstacles and other vehicles, thereby generating a wide range of driving scenario datasets.

[0059] To enhance the realism of the simulation results, the system incorporates a large amount of real human driving data. This data covers driver behavior patterns under different curve conditions, such as deceleration, steering, and acceleration. This behavioral data is embedded into the Monte Carlo simulation, making the risk assessment closer to actual human driving behavior, thereby improving the accuracy of predicting real-world driving risks.

[0060] Based on scenario and risk data generated from Monte Carlo simulations, the system performs statistical analysis on all possible risk values ​​to form a risk distribution model. By analyzing this model, the system determines a reasonable collision risk threshold. This threshold is used to determine whether a collision warning needs to be triggered during actual driving when the real-time calculated risk value exceeds this threshold.

[0061] During vehicle operation, the system calculates the current risk value in real time and compares it with a preset collision risk threshold. When the calculated risk value exceeds the set threshold, the system immediately issues a collision warning signal to avoid a potential collision.

[0062] In this embodiment, through multi-sensor data fusion and processing, the system acquires real-time vehicle status information such as speed, position, steering wheel angle, heading angle, and acceleration. Simultaneously, it acquires environmental information such as the speed, position, and historical trajectories of nearby vehicles, as well as road curvature, width, and road surface conditions of curves. This data is transmitted to the EC via a 5G network, and a Bi-LSTM model is used to predict the driving trajectories of non-networked vehicles.

[0063] In this embodiment, the system establishes a driving risk field model using a Gaussian function and dynamically adjusts the model parameters based on real-time data to accurately assess the collision risk of the vehicle in curves. A vehicle curve model can be constructed based on current driving parameters, and the turning radius of the vehicle when turning in the current curve can be analyzed based on the vehicle curve model. The vehicle curve model can be referenced from... Figure 2 When a risk value exceeds a set threshold, the system immediately issues a warning signal to avoid potential collisions. This rapid response mechanism significantly improves vehicle safety in complex curves.

[0064] A method and system for warning of cornering collisions in connected and cooperative driving vehicles, such as Figure 3 As shown, the system includes: a status information acquisition unit, a trajectory prediction unit, a risk assessment module, and a collision warning module;

[0065] The status information acquisition unit is used to acquire the vehicle's status information, the status information of surrounding vehicles, and road environment information. The status information acquisition unit uses multiple sensors such as high-precision GPS, lidar, millimeter-wave radar, and cameras to collect information such as the vehicle's speed, position, and direction in real time, while also acquiring dynamic data of nearby vehicles and environmental information such as road curvature, width, and road surface conditions of curves.

[0066] The trajectory prediction unit predicts the driving trajectories of non-connected vehicles around the vehicle based on the input information from the information acquisition unit; the trajectory prediction unit can identify potential collision risks in advance, providing important predictive data support for subsequent risk assessment.

[0067] The risk assessment module calculates the risk of a vehicle's cornering collision in real time based on the vehicle's information, surrounding vehicle information, and road environment information; the risk assessment module ensures the comprehensiveness and accuracy of the risk calculation and provides the system with real-time risk status information.

[0068] The collision avoidance warning module compares the real-time collision risk of the vehicle on the curve with a built-in collision risk threshold. When the risk exceeds the threshold, a collision risk warning is issued. Through real-time monitoring and rapid response, the collision avoidance warning module effectively improves the safety of the vehicle when driving on curves.

[0069] In this embodiment, the system includes at least: an RSU for collecting driving data from non-connected vehicles; an edge cloud processing module for fusing and preprocessing the input data, and using a Bi-LSTM model to predict the driving trajectory of non-connected vehicles at curves; a driving risk field model for combining the driving risk field model of intelligent connected vehicles to assess the collision risk of driving on curves in real time; and a collision warning system for calculating a collision risk threshold using the Monte Carlo method combined with human driving data, and issuing a collision warning when the real-time calculated risk value exceeds the set threshold.

[0070] In one specific embodiment, the RSU comprehensively collects driving data (such as speed, position, and steering wheel angle) and surrounding environmental information (such as road curvature and lane ID) from non-connected vehicles using multiple sensors, including cameras, LiDAR, and high-precision GPS. This data is transmitted to the edge cloud in real time via a 5G communication network. The edge cloud fuses and preprocesses the received data, using a Bi-LSTM model to predict the vehicle's trajectory. The predicted trajectory data is further input into a driving risk field model to dynamically assess collision risks in curves, ensuring that each assessment is based on the latest vehicle motion prediction results. When the real-time risk value exceeds a preset threshold, the system immediately issues a warning signal via the 5G network. The system adopts a modular design, supporting future upgrades or integration with other advanced driver assistance systems, thereby continuously improving driving safety.

[0071] The system implementation method of the present invention is the same as the method implementation method.

[0072] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include ROM, RAM, disk, or optical disk, etc.

[0073] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for curve collision warning of a connected cooperative driving vehicle, characterized in that, include: S1. The roadside RSU collects vehicle status information and curve environment information and uploads it to the EC, where EC is an edge cloud server; S2 and EC fuse the input data and use a Bi-LSTM model to predict the driving trajectory of non-connected vehicles at curves; S3. Establish a driving risk field model for intelligent connected vehicles, and combine it with the predicted trajectories of surrounding vehicles to assess the collision risk of driving on curves in real time. Establish a driving risk field model for intelligent connected vehicles, including: Where R(x,y) is the collision risk function for point (x,y), h represents the height of the Gaussian cross section function, w represents the width of the Gaussian cross section function, and r c Represents the vehicle's turning radius, (x o ,y o () represents the coordinates of the center point of the turning radius; S4. Utilize the Monte Carlo method and combine it with human driving data to calculate the collision risk threshold and provide cornering collision warning. Using the Monte Carlo method and combining human driving data to calculate collision risk thresholds, a cornering collision warning system is implemented, including: The collision risk threshold is calculated using the Monte Carlo method, corrected using human driving data, and then compared with the real-time collision risk to complete the cornering collision warning. The Monte Carlo method is used to calculate the collision risk threshold, including: Among them, R k represents the driving collision risk value, meshgrids represents the set of grid points, and Sc represents the driving scenario cost of intelligent connected vehicles.

2. The method for curve collision warning of a connected cooperative driving vehicle according to claim 1, characterized in that, The information collected by the roadside RSU includes: the vehicle's status information, the status information of surrounding vehicles, and the curve environment information; The vehicle's status information includes: vehicle position, vehicle steering wheel angle, vehicle heading angle, vehicle speed, and vehicle dimensions; The status information of the surrounding vehicles includes: vehicle position, vehicle speed, vehicle historical trajectory, and vehicle size; The curve environment information includes: lane curvature and lane ID.

3. The method for curve collision warning of a connected cooperative driving vehicle according to claim 1, characterized in that, EC integrates the input data, including: By fusing EC data, the input information for the non-connected vehicle trajectory prediction model is determined.

4. The method for curve collision warning of a connected cooperative driving vehicle according to claim 1, characterized in that, The Bi-LSTM model comprises two independent LSTM networks, used to process the forward and reverse information of the input sequence, respectively, to predict the driving trajectory of non-networked vehicles at curves. The forward LSTM processes the sequence from x1 to x... t And generate a hidden state at each time step t. Inverse LSTM processing sequence from x t To x1, and generate a hidden state at each time step t. For each time step t, the hidden states generated by the forward and backward LSTMs are... and The parts are concatenated together to form a single feature vector h. t .

5. A method for curve collision warning of a connected cooperative driving vehicle according to claim 4, characterized in that, The hidden states generated by forward and backward LSTM and The parts are concatenated together to form a single feature vector h. t , include: Here, α represents the contribution weight parameter used to control the positive and negative feature vectors during merging.

6. A curve collision warning system for a connected and cooperative driving vehicle, the system being used to execute a curve collision warning method for a connected and cooperative driving vehicle as described in any one of claims 1 to 5, characterized in that, include: The system includes a status information acquisition unit, a trajectory prediction unit, a risk assessment module, and a collision warning module. The status information acquisition unit is used to acquire the vehicle's status information, the status information of surrounding vehicles, and road environment information; The trajectory prediction unit predicts the driving trajectories of non-networked vehicles around the vehicle based on the input information from the status information acquisition unit. The risk assessment module calculates the risk of the vehicle's cornering collision in real time based on the vehicle's information, surrounding vehicle information, and road environment information. The collision warning module compares the real-time collision risk of the vehicle at the curve with a built-in collision risk threshold, and issues a collision risk warning when the risk exceeds the threshold.

Images