Vehicle collision early warning method based on driving sight distance

Abstract

The invention discloses a vehicle collision early warning method based on a driving sight distance. The method comprises the following steps: step 1, a target vehicle divides a front road early warning coverage distance into a plurality of dynamic virtual sight distance sections; 2, selecting one vehicle from each dynamic virtual sight distance section as an agent node; step 3, the proxy node collects traffic target data in the dynamic virtual sight distance section, compresses the traffic target data into summary information and transmits the summary information to the edge node; the edge node fuses the summary information transmitted by each agent node to construct a global dynamic risk map; and step 4, the edge node predicts the collision probability of the target vehicle in each dynamic virtual sight distance section based on the global dynamic risk map, generates a graded early warning instruction, transmits the graded early warning instruction to each agent node, and notifies the target vehicle through the agent nodes. According to the invention, complex road scenes can be accurately covered, the node reliability is enhanced, the bottleneck of communication and computing power is broken through, and reliable guarantee is provided for the safety of the Internet of Vehicles.

Description

Technical Field

[0001] This invention relates to the field of vehicle collision warning methods, specifically a vehicle collision warning method based on driving sight distance. Background Technology

[0002] Traffic sight distance (DSD) warning sharing refers to the real-time sharing and exchange of DSD data (the distance between the vehicle and a target vehicle or obstacle ahead) and related warning information among multiple vehicles in motion via wireless communication or other means. In this way, each vehicle can not only obtain forward information using its own sensors but also acquire DSD and warning data from other vehicles, achieving multi-vehicle collaboration and information complementarity, thereby improving overall road traffic safety and warning accuracy. Existing collision warning methods based on DSD include two types: static threshold methods based on single-vehicle sensors and probabilistic prediction methods based on multi-sensor fusion.

[0003] Static threshold methods based on single-vehicle sensors rely on a single sensor such as onboard radar or a camera to monitor parameters such as the distance and relative speed of targets ahead (vehicles, pedestrians, etc.) in real time. A warning is triggered when the target distance is less than a preset safety threshold (e.g., a fixed distance formula based on vehicle speed, such as "vehicle speed × reaction time + braking distance"). For example, the forward collision warning function in traditional adaptive cruise control (ACC) systems monitors the distance to the vehicle ahead using millimeter-wave radar and issues an alarm when the distance is less than a set threshold. This method relies solely on single-vehicle perception, without incorporating vehicle-to-vehicle communication, and the safety threshold is statically set, making it unsuitable for adapting to complex road conditions and environmental changes.

[0004] This probabilistic prediction method, based on multi-sensor fusion, integrates data from multiple sensors such as radar, cameras, and lidar. It predicts the future trajectory of a target using kinematic models (e.g., constant velocity or constant acceleration models) and calculates the time difference (TTC, time of collision) between the two vehicle trajectories. A warning is triggered when the TTC is less than a preset threshold. While this method improves perception accuracy, it still relies primarily on single-vehicle perception and does not utilize multi-vehicle collaborative data. Furthermore, the trajectory prediction does not consider macroscopic road conditions such as road curvature, limiting its accuracy in complex scenarios.

[0005] The following defects exist in the existing technology:

[0006] Rigid spatial division: The use of fixed-length segments cannot adapt to changes in sight distance on complex roads such as curves and slopes, resulting in insufficient blind spot coverage and high warning delay in high-speed scenarios;

[0007] Low node reliability: Relying on signal strength or a simple voting mechanism to elect nodes, ignoring positioning drift and sensor calibration errors, resulting in a high false alarm rate;

[0008] Communication efficiency bottlenecks: The transmission of raw point clouds / images consumes too much bandwidth, channels are congested in high-density areas, and static warning thresholds cannot adapt to rain, fog, or nighttime environments, leading to a surge in false alarms and missed alarms. Summary of the Invention

[0009] This invention provides a vehicle collision warning method based on driving sight distance to solve the problem of rigid spatial division in the prior art.

[0010] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0011] The vehicle collision warning method based on driving sight distance proceeds as follows:

[0012] Step 1: Take one of the vehicles on the road as the target vehicle, and divide the warning coverage distance of the road ahead into several dynamic virtual line-of-sight segments by the target vehicle, thereby constructing a dynamic virtual line-of-sight chain.

[0013] Step 2: Select one vehicle from each vehicle in each dynamic virtual line-of-sight segment as the proxy node;

[0014] Step 3: Each agent node collects traffic target data within its own dynamic virtual line-of-sight segment and compresses it into summary information, and each agent node transmits the summary information to the edge node;

[0015] The edge nodes fuse the summary information transmitted by the proxy nodes in each dynamic virtual line-of-sight segment, thereby constructing a global dynamic risk map;

[0016] Step 4: The edge nodes predict the collision probability of the target vehicle in each dynamic virtual line-of-sight segment based on the global dynamic risk map, and generate graded warning instructions which are transmitted to the agent nodes of each dynamic virtual line-of-sight segment. The agent nodes then distribute the warnings to the target vehicles entering their respective dynamic virtual line-of-sight segments.

[0017] Furthermore, in step 1, the target vehicle, based on the road centerline coordinate sequence of the high-precision map, calculates the actual length of each virtual line-of-sight segment in the forward road warning coverage distance sequentially along the driving direction, starting from the target vehicle's own position.

[0018] When calculating the actual length of each virtual line-of-sight segment, the target vehicle obtains the curvature and average speed of the road ahead of the target vehicle through high-precision maps and real-time perception, and calculates the curvature correction coefficient of the road ahead of the target vehicle; then, the target vehicle calculates the actual length of each virtual line-of-sight segment based on the preset length of the virtual line-of-sight segment, the preset safe time threshold, the curvature, the average speed, and the curvature correction coefficient.

[0019] The actual length of all virtual line-of-sight segments in the warning coverage distance of the road ahead is calculated sequentially by the target vehicle until the warning coverage distance of the road ahead is completely divided, thus obtaining a dynamic virtual line-of-sight chain composed of several consecutive dynamic virtual line-of-sight segments.

[0020] Furthermore, in step 1, the actual length of each virtual line-of-sight segment is calculated using the following formula:

[0021]

[0022] Where: L seg The actual length of the virtual line-of-sight segment; L max Indicates the preset length of the virtual line-of-sight segment; V agv T represents the average speed of the vehicle ahead on the road; safe α represents the preset safe time threshold; α represents the calculated curvature correction coefficient; κ represents the curvature of the road in front of the target vehicle.

[0023] Furthermore, in step 2, the positioning error coefficient, sensor error coefficient, and communication quality coefficient of each vehicle in each dynamic virtual line-of-sight segment are substituted into the proxy node election model to calculate the election coefficient of each vehicle; then the vehicle with the largest election coefficient in each dynamic virtual line-of-sight segment is selected as the proxy node in the corresponding dynamic virtual line-of-sight segment.

[0024] Furthermore, in step 2, the calculation formula for the proxy node election model is as follows:

[0025]

[0026] in: This represents the election coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; This represents the positioning error coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; This represents the sensor error coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; φ1 represents the communication quality coefficient of vehicle a in the i-th dynamic virtual line-of-sight segment; φ1, φ2, and φ3 represent the preset positioning error weighting factor, sensor error weighting factor, and communication quality weighting factor, respectively; i represents the number of the dynamic virtual line-of-sight segment, i = 1, 2, ..., j, where j is a positive integer greater than 2; a represents the number of the vehicle in the dynamic virtual line-of-sight segment, a = 1, 2, ..., b, where b is a positive integer greater than 2; e represents a natural constant.

[0027] Furthermore, in step 3, the traffic target data collected by the agent node within its dynamic virtual line-of-sight segment includes dynamic target type and quantity, key motion parameters, and abnormal event flags. The agent node compresses the dynamic target type and quantity, key motion parameters, and abnormal event flags into summary information and transmits it to the edge node.

[0028] Furthermore, in step 3, the process of the edge nodes constructing a global dynamic risk map is as follows:

[0029] Step 3.1) Data Collection and Preprocessing

[0030] Edge nodes periodically receive summary information uploaded by proxy nodes in each dynamic virtual line-of-sight segment, and perform time synchronization and spatial registration of the summary information to ensure that all data is mapped to a unified map road coordinate system and time axis.

[0031] Step 3.2) Spatial Mapping and Partitioning

[0032] Edge nodes spatially partition roads on the map according to dynamic virtual line-of-sight segments, so that each dynamic virtual line-of-sight segment partition corresponds to a different spatial unit on the map.

[0033] Step 3.3) Risk Factor Calculation

[0034] For each dynamic virtual line-of-sight segment, the edge nodes calculate the risk factor of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on the summary information;

[0035] Step 3.4) Map fusion and dynamic updating

[0036] Edge nodes map the risk factors of all dynamic virtual line-of-sight segments onto the global road map, forming a one-dimensional or two-dimensional global dynamic risk map.

[0037] Furthermore, in step 4, the edge node predicts the future trajectory of all dynamic targets in each dynamic virtual line-of-sight segment; when the predicted future trajectory is linear, the edge node uses a machine learning model to predict the collision probability of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on the risk factor; when the predicted future trajectory is non-linear, the edge node uses an LSTM deep learning model to output the collision probability of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on the risk factor.

[0038] Finally, by combining the environmental credibility factor at the edge nodes, the predicted collision probability of the target vehicle in each dynamic virtual line-of-sight segment is calculated.

[0039] Furthermore, in step 4, the edge node compares the predicted collision probability of the target vehicle in each dynamic virtual line-of-sight segment with the preset predicted collision probability ranges for non-warning level, prompt level, warning level, and emergency level, respectively.

[0040] If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset non-warning level predicted collision probability range, then the warning level of the target vehicle in that dynamic virtual line-of-sight segment is determined to be non-warning level.

[0041] If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset warning level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the warning level.

[0042] If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset warning level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the warning level.

[0043] If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset emergency level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the emergency level.

[0044] Compared with the prior art, the advantages of the present invention are:

[0045] 1. Precise coverage of complex road scenarios: Based on road curvature, slope and real-time vehicle speed, virtual sight distance segments are dynamically divided, which significantly improves the warning coverage of blind spots such as curves and slopes, and ensures that no area with limited sight distance is missed;

[0046] 2. Enhance node reliability: A three-dimensional election model that integrates positioning error compensation, sensor calibration residuals, and communication quality accurately selects highly reliable proxy nodes, significantly reducing the risk of false alarms and missed alarms;

[0047] 3. Overcoming communication and computing power bottlenecks: The original data transmission is replaced by hierarchical summary information (target distribution histogram, key motion parameters and event flags), combined with intelligent compression technology, which efficiently utilizes communication resources and reduces the on-board computing load.

[0048] 4. The overall solution forms a technical closed loop of "dynamic perception - trusted collaboration - lightweight early warning", which significantly improves the real-time performance and system robustness of early warning in complex traffic environments and provides reliable protection for vehicle network security. Attached Figure Description

[0049] Figure 1 This is a flowchart of the method according to an embodiment of the present invention. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] like Figure 1 As shown in the figure, this embodiment discloses a vehicle collision warning method based on driving sight distance, the process of which is as follows:

[0052] Step 1: Using one of the vehicles on the road as the target vehicle, the target vehicle divides the warning coverage distance of the road ahead into several dynamic virtual line-of-sight segments, thereby constructing a dynamic virtual line-of-sight chain.

[0053] In this embodiment, the target vehicle uses the road centerline coordinate sequence of the high-precision map as the starting point and calculates the actual length of each virtual line-of-sight segment in the forward road warning coverage distance along the driving direction.

[0054] When calculating the actual length of each virtual line-of-sight segment, the target vehicle obtains the curvature and average speed of the road ahead of it through high-precision maps and real-time perception, and calculates the curvature correction coefficient of the road ahead. Then, based on the preset length of the virtual line-of-sight segment, the preset safe time threshold, the curvature, the average speed, and the curvature correction coefficient, the target vehicle calculates the actual length of each virtual line-of-sight segment. The formula for calculating the actual length of each virtual line-of-sight segment is as follows:

[0055]

[0056] Where: L seg The actual length of the virtual line-of-sight segment; L max Indicates the preset length of the virtual line-of-sight segment; V agv T represents the average speed of the vehicle ahead on the road; safe α represents the preset safe time threshold; α represents the calculated curvature correction coefficient; κ represents the curvature of the road in front of the target vehicle.

[0057] In this embodiment, the formula for calculating the curvature correction coefficient of the road in front of the target vehicle is:

[0058]

[0059] Where: A represents the basic curvature sensitivity coefficient, and B represents the vehicle speed attenuation adjustment factor.

[0060] In this embodiment, the basic curvature sensitivity coefficient A is the benchmark value for the curvature influence weight in the dynamic segmentation formula, characterizing the adjustment strength of a unit curvature change on the virtual sight distance segment length in low-speed scenarios. The basic curvature sensitivity coefficient A is essentially a balance factor between road geometric risk and safety warning sensitivity. It is obtained through statistical analysis of real vehicle test data and set by professionals; in this embodiment, A = 0.25.

[0061] In this embodiment, the vehicle speed attenuation adjustment factor B is a key parameter in the dynamic segmented formula that quantifies the weakening effect of vehicle speed on curvature sensitivity. It is obtained through real vehicle tests at multiple speed ranges, fitting with an exponential model, and finally verifying the safety boundary. It is set by professionals. In this embodiment, B = 0.012.

[0062] In this embodiment, the preset length L of the virtual line-of-sight segment is obtained from the database. max This embodiment, based on road design specifications and historical accident data, sets differentiated initial preset lengths according to road type (highway / urban / mountainous), introduces real-time vehicle speed and curvature ratio factors, and dynamically scales the segment length based on design speed and standard curvature (the higher the vehicle speed, the longer the segment length; the greater the curvature, the shorter the segment length). This is set by professionals. In this embodiment, L... max It is 200 meters.

[0063] In this embodiment, a preset security time threshold T is obtained from the database. safe The preset safety time threshold T safe The settings are configured by professionals. For example, a baseline value is first initialized based on road type (3.0s for highways, 2.5s for cities, and 4.0s for mountainous areas). Then, real-time visibility, rainfall intensity, and traffic density are dynamically integrated for linear weighted correction (thresholds are increased by 30%-50% in dense fog and rainstorm scenarios and compressed by 20% during congestion). This embodiment also establishes an automatic database update mechanism, triggering threshold recalculation after extreme weather events or when accident rate fluctuations exceed 10%, ensuring that the safety margin always matches the actual risk level.

[0064] Finally, in this embodiment, the actual length of all virtual line-of-sight segments in the warning coverage distance of the road ahead is calculated sequentially by the target vehicle until the warning coverage distance of the road ahead is completely divided, thereby obtaining a dynamic virtual line-of-sight chain composed of several continuous dynamic virtual line-of-sight segments.

[0065] Step 2: Select one vehicle from all vehicles in each dynamic virtual line-of-sight segment to serve as the proxy node for the corresponding dynamic virtual line-of-sight segment.

[0066] In this embodiment, the positioning error coefficient, sensor error coefficient, and communication quality coefficient of each vehicle in each dynamic virtual line-of-sight segment are obtained. These coefficients are then substituted into the proxy node election model to calculate the election coefficient for each vehicle in the corresponding dynamic virtual line-of-sight segment. Finally, the vehicle with the highest election coefficient in each dynamic virtual line-of-sight segment is selected as the proxy node for that segment.

[0067] In this embodiment, the calculation formula for the proxy node election model is as follows:

[0068]

[0069] in: This represents the election coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; This represents the positioning error coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; This represents the sensor error coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; φ1 represents the communication quality coefficient of vehicle a in the i-th dynamic virtual line-of-sight segment; φ1, φ2, and φ3 represent the preset positioning error weighting factor, sensor error weighting factor, and communication quality weighting factor, respectively; i represents the number of the dynamic virtual line-of-sight segment, i = 1, 2, ..., j, where j is a positive integer greater than 2; a represents the number of the vehicle in the dynamic virtual line-of-sight segment, a = 1, 2, ..., b, where b is a positive integer greater than 2; e represents a natural constant.

[0070] In this embodiment, the positioning error coefficient is used to quantify the uncertainty of the vehicle's absolute position, including the combined effects of GNSS positioning deviation and IMU cumulative drift. The sensor error coefficient is used to characterize the degree of deviation between sensor measurements and the true values. The communication quality coefficient is used to evaluate the transmission reliability of the V2X link, fusing packet loss rate and signal strength.

[0071] Specifically, in this embodiment, the positioning error coefficient is calculated using the GNSS / RTK positioning residual and the IMU drift compensation. The IMU drift compensation is calculated using an established heading angle drift model, the expression of which is:

[0072]

[0073] Where: Δθ(t) represents the IMU heading angle drift; ω bias β represents the gyroscope's zero bias; β and γ represent the calibration coefficients; t represents the system's runtime.

[0074] It should be noted that gyroscope zero bias is the non-zero angular velocity value output by the gyroscope when there is no rotational input; it is essentially an inherent systematic error of the sensor. Calibration coefficients are the key to decoding the sensor error chain. β reveals the propagation gain of the systematic error, and γ quantifies the accumulation rate of random noise. Together, they construct the mapping relationship of "physical error → mathematical model," providing the theoretical basis for error compensation in high-precision navigation.

[0075] The positioning error coefficient P is calculated by combining the GNSS / RTK positioning residual and the IMU drift compensation calculated from the heading angle drift model. loc As shown in the following formula:

[0076]

[0077] Where: σ gps This represents the raw GNSS error, and D represents the distance the vehicle traveled.

[0078] In this embodiment, if the sensor is a camera, the image distortion correction residual (such as reprojection error) is used as the sensor error coefficient E. sensor The calculation formula is:

[0079]

[0080] Where: p m Represents the actual pixel; This represents the corrected pixel value; n is the number of sampling points; m represents the sampling point number.

[0081] If the sensor is a lidar, then the mean square error of point cloud registration (MSE) is used as the sensor error coefficient E. sensor The calculation formula is:

[0082]

[0083] Where: P N Represents the point cloud of the current frame; Q N This represents the reference frame point cloud; N is the number of point clouds; M represents the point cloud number.

[0084] If the sensor is a multi-sensor fusion, then a weighted average can be used to calculate the sensor error coefficient E. sensor The calculation formula is:

[0085]

[0086] Among them: E sensor,k λ represents the sensor error coefficient of the k-th sensor. k This represents the sensor weight of the k-th sensor; k represents the sensor number.

[0087] The communication quality coefficient Q in this embodiment comm The calculation formula is as follows:

[0088]

[0089] Wherein: PLR represents packet loss rate, which can be obtained through statistics from the communication module; SNR represents signal-to-noise ratio, which can be measured in real time through the communication module.

[0090] In this embodiment, the preset positioning error weighting factor, sensor error weighting factor, and communication quality weighting factor are all set by professionals and stored in a database. The system employs a three-mechanism approach: inverse accuracy weighting (positioning / communication), error source decomposition (sensor), and dynamic amplification based on environmental perception. These three mechanisms are all existing technologies and will not be elaborated upon further here.

[0091] Step 3: Each agent node collects traffic target data within its respective dynamic virtual line-of-sight segment and compresses it into summary information. Each agent node then transmits this summary information to the edge nodes. The edge nodes fuse the summary information transmitted by the agent nodes in each dynamic virtual line-of-sight segment, perform analysis and spatial mapping, thereby constructing a global dynamic risk map covering the road area and reflecting the distribution and dynamic changes of traffic risks.

[0092] In this embodiment, the traffic target data collected by the agent node within its dynamic virtual line-of-sight segment includes dynamic target types and quantities, key motion parameters, and abnormal event flags. Target types include, but are not limited to, vehicles, pedestrians, non-motorized vehicles, and static obstacles. Key motion parameters include, but are not limited to, maximum relative speed and average density. Maximum relative speed refers to the maximum relative speed of other vehicles relative to the target vehicle, and average density refers to the total number of all target types in each dynamic virtual line-of-sight segment. Abnormal event flags include, but are not limited to, accident flags, rockfall flags, and pothole flags. The agent node compresses the dynamic target types and quantities, key motion parameters, and abnormal event flags into summary information and transmits it to the edge nodes.

[0093] In this embodiment, the process of edge nodes constructing a global dynamic risk map is as follows:

[0094] Step 3.1) Data Collection and Preprocessing

[0095] Edge nodes periodically receive summary information uploaded by proxy nodes in each dynamic virtual line-of-sight segment, and perform time synchronization and spatial registration of the summary information to ensure that all data is mapped to a unified map road coordinate system and time axis.

[0096] Step 3.2) Spatial Mapping and Partitioning

[0097] Edge nodes spatially partition roads on the map according to dynamic virtual view distance segments, so that each dynamic virtual view distance segment partition corresponds to a different spatial unit on the map, which can be a line segment, a grid, or a polygon.

[0098] Step 3.3) Risk Factor Calculation

[0099] For each dynamic virtual line-of-sight segment, the edge nodes calculate the risk factor of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on the summary information; and historical data and environmental information (such as visibility and weather) are introduced to further correct the risk factor.

[0100] In this embodiment, the edge nodes calculate the risk factor of the target vehicle in each dynamic virtual line-of-sight segment based on the summary information of each segment. The calculation formula is as follows:

[0101]

[0102] Among them: Risk_Score i N represents the risk factor for the i-th dynamic virtual line-of-sight segment; target,i L represents the total number of dynamic targets in the i-th dynamic virtual line-of-sight segment; seg,i V represents the actual length of the i-th dynamic virtual line-of-sight segment; relmax,i ρ represents the maximum relative velocity in the i-th dynamic virtual line-of-sight segment; avg,i Represents the average density of the i-th dynamic virtual view segment; Event flag,i This represents the risk coefficient corresponding to the abnormal event flag in the i-th dynamic virtual line-of-sight segment; These represent the preset target total weight factor, maximum relative speed weight factor, average density weight factor, and risk coefficient weight factor corresponding to the abnormal event flag in the database, respectively.

[0103] It should be noted that the preset weighting factors for the total number of targets, maximum relative speed, average density, and risk coefficients corresponding to abnormal event markers in the database are set by professionals. For example, the weighting principle is: dynamically allocated based on business priority (e.g., speed > density), data distribution characteristics, and risk sensitivity, with the total weight always being 1.

[0104] Furthermore, in this embodiment, the rationality of the weights is verified by backtracking through historical data, the risk coefficient weight factor of the weight factors is optimized by using sensitivity analysis (such as Morris method), and a scenario-based dynamic adjustment mechanism is established (such as automatically increasing the speed weight in extreme weather).

[0105] In the risk factor calculation formula, the factors are treated differently: the dynamic target number is calculated using a piecewise function (marginal effect decay), the velocity is amplified exponentially according to the square relationship, the density is calculated according to the deviation of the normal distribution, and abnormal events are significantly weighted as switch variables.

[0106] Step 3.4) Map fusion and dynamic updating

[0107] Edge nodes map the risk factors of all dynamic virtual line-of-sight segments onto the global road map, forming a one-dimensional or two-dimensional global dynamic risk map. For overlapping or boundary areas, a weighted average or maximum value fusion strategy is used. The map is dynamically refreshed as the summary information is updated in real time, reflecting the spatiotemporal changes in traffic risks.

[0108] Step 4: The edge nodes predict the collision probability of the target vehicle in each dynamic virtual line-of-sight segment based on the global dynamic risk map, and generate graded warning instructions which are transmitted to the agent nodes of each dynamic virtual line-of-sight segment. The agent nodes then distribute the warnings to the target vehicles entering their respective dynamic virtual line-of-sight segments.

[0109] In this embodiment, the process by which edge nodes predict the collision probability of the target vehicle in each dynamic virtual line-of-sight segment is as follows:

[0110] Edge nodes acquire all dynamic targets of the target vehicle in each dynamic virtual line-of-sight segment of the global dynamic risk map, and predict the future trajectory of all dynamic targets in each dynamic virtual line-of-sight segment.

[0111] When the predicted future trajectory is linear, the edge node uses a machine learning model to predict the collision probability of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on risk factors; when the predicted future trajectory is nonlinear, the edge node uses an LSTM deep learning model to predict the collision probability of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on risk factors.

[0112] The machine learning model predicts as shown in the following equation:

[0113] [P i,tra =f1(Δd,Δv,Risk_Score) i )],

[0114] Where: Δd represents the predicted nearest distance; Δv represents the relative velocity; and f1() represents the machine learning model.

[0115] The predictions made by the LSTM deep learning model are shown in the following equation:

[0116] [P i,lstm =f2(historical trajectory, Risk_Score) i )],

[0117] Where f2() represents the LSTM deep learning model.

[0118] In this final embodiment, the edge node integrates the environmental confidence factor and uses this factor to correct the predicted collision probability, thus calculating the predicted collision probability of the target vehicle in each dynamic virtual line-of-sight segment. The calculation formula is as follows:

[0119]

[0120] Where: η represents the environmental credibility factor; P i This represents the predicted collision probability of the target vehicle in the i-th dynamic virtual line-of-sight segment.

[0121]

[0122] Where R rain N represents rainfall intensity. light The formula parameters (such as 0.2, 0.7, 0.3, 0.05, 0.9, 0.1, etc.) representing street light coverage are set based on the influence of communication quality, rainfall, and illumination on perception credibility in actual scenarios, combined with regression analysis of a large amount of experimental data and engineering experience: the communication quality factor uses an attenuation coefficient of 0.2 to control the degree of credibility decay when communication is interrupted; the rainfall intensity factor uses 0.7 as the lower limit of credibility during heavy rainfall, and the exponential coefficients of 0.3 and 0.05 fit the exponential decay law of perception credibility with rainfall intensity; the street light coverage factor uses 0.9 as the basic credibility when there are no street lights, and the gain coefficient of 0.1 reflects the credibility improvement when there is sufficient illumination. Ultimately, the environmental credibility factor can objectively quantify the impact of the environment on perception, providing reliable weights for decision-making.

[0123] This embodiment uses the environmental credibility factor as a key parameter to quantify the reliability of sensing data. Its calculation requires a comprehensive consideration of sensor performance, environmental interference, and target feature uncertainty. It is calculated through a physical model and statistical rules, which is existing technology and will not be elaborated on here.

[0124] In this embodiment, the edge node compares the predicted collision probability of the target vehicle in each dynamic virtual line-of-sight segment with the preset predicted collision probability ranges for non-warning level, prompt level, warning level, and emergency level, respectively.

[0125] If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset non-warning level predicted collision probability range, then the warning level of the target vehicle in that dynamic virtual line-of-sight segment is determined to be non-warning level.

[0126] If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset warning level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the warning level.

[0127] If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset warning level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the warning level.

[0128] If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset emergency level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the emergency level.

[0129] It should be noted that in this embodiment, the predicted collision probability ranges for non-warning level, prompt level, warning level, and emergency level are preset in the database. These ranges are all set by professionals. For example:

[0130] Baseline setting: Set an initial threshold based on system response time (such as braking time) (emergency level ≥ 0.75).

[0131] By scenario: dynamically adjusted according to road type / environment (emergency level for highways is reduced to 0.65, and the warning range is compressed in rainy and foggy weather).

[0132] Physical verification: Verify that the threshold conforms to kinematic constraints (a warning level must allow ≥1.5 seconds of intervention time).

[0133] In this embodiment, the agent node distributes warnings to target vehicles entering the dynamic virtual line-of-sight segment. The specific implementation method is as follows: through vehicle-to-vehicle communication, vehicle-to-infrastructure communication, and V2V and V2I hybrid communication, the graded warning instructions are distributed to the target vehicles entering the dynamic virtual line-of-sight segment. When distributing, geofence broadcasting, targeted propagation, and information priority strategies are adopted.

[0134] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. These embodiments are merely descriptions of preferred embodiments and are not intended to limit the scope or concept of the invention. The specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. Such combinations, as long as they do not violate the spirit of the present invention, should also be considered as part of this disclosure. To avoid unnecessary repetition, the present invention will not further describe the various possible combinations.

[0135] This invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this invention and without departing from the design idea of ​​this invention, all modifications and improvements made by those skilled in the art to the technical solutions of this invention should fall within the protection scope of this invention. The technical content for which protection is sought in this invention has been fully described in the claims.

Claims

1. A vehicle collision warning method based on driving sight distance, characterized in that, The process is as follows: Step 1: Take one of the vehicles on the road as the target vehicle, and divide the warning coverage distance of the road ahead into several dynamic virtual line-of-sight segments by the target vehicle, thereby constructing a dynamic virtual line-of-sight chain. Step 2: Select one vehicle from each vehicle in each dynamic virtual line-of-sight segment as the proxy node; Step 3: Each agent node collects traffic target data within its own dynamic virtual line-of-sight segment and compresses it into summary information, and each agent node transmits the summary information to the edge node; The edge nodes fuse the summary information transmitted by the proxy nodes in each dynamic virtual line-of-sight segment, thereby constructing a global dynamic risk map; Step 4: The edge nodes predict the collision probability of the target vehicle in each dynamic virtual line-of-sight segment based on the global dynamic risk map, and generate graded warning instructions which are transmitted to the agent nodes of each dynamic virtual line-of-sight segment. The agent nodes then distribute the warnings to the target vehicles entering their respective dynamic virtual line-of-sight segments.

2. The vehicle collision warning method based on driving sight distance according to claim 1, characterized in that, In step 1, the target vehicle uses the road centerline coordinate sequence of the high-precision map as the starting point and calculates the actual length of each virtual line-of-sight segment in the forward road warning coverage distance along the driving direction. When calculating the actual length of each virtual line-of-sight segment, the target vehicle obtains the curvature and average speed of the road in front of the target vehicle through high-precision maps and real-time perception, and calculates the curvature correction coefficient of the road in front of the target vehicle. Then, the target vehicle calculates the actual length of each virtual line-of-sight segment based on the preset length of the virtual line-of-sight segment, the preset safe time threshold, the curvature, the average vehicle speed, and the curvature correction coefficient. The actual length of all virtual line-of-sight segments in the warning coverage distance of the road ahead is calculated sequentially by the target vehicle until the warning coverage distance of the road ahead is completely divided, thus obtaining a dynamic virtual line-of-sight chain composed of several consecutive dynamic virtual line-of-sight segments.

3. The vehicle collision warning method based on driving sight distance according to claim 2, characterized in that, In step 1, the formula for calculating the actual length of each virtual line-of-sight segment is as follows: Where: L seg The actual length of the virtual line-of-sight segment; L max Indicates the preset length of the virtual line-of-sight segment; V agv T represents the average speed of the vehicle ahead on the road; safe α represents the preset safe time threshold; α represents the calculated curvature correction coefficient; κ represents the curvature of the road in front of the target vehicle.

4. The vehicle collision warning method based on driving sight distance according to claim 1, characterized in that, In step 2, the positioning error coefficient, sensor error coefficient, and communication quality coefficient of each vehicle in each dynamic virtual line-of-sight segment are substituted into the proxy node election model to calculate the election coefficient of each vehicle. Then, the vehicle with the largest election coefficient in each dynamic virtual line-of-sight segment is selected as the proxy node in the corresponding dynamic virtual line-of-sight segment.

5. The vehicle collision warning method based on driving sight distance according to claim 4, characterized in that, In step 2, the calculation formula for the proxy node election model is as follows: in: This represents the election coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; This represents the positioning error coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; This represents the sensor error coefficient of the a-th vehicle in the i-th dynamic virtual line-of-sight segment; φ1 represents the communication quality coefficient of vehicle a in the i-th dynamic virtual line-of-sight segment; φ1, φ2, and φ3 represent the preset positioning error weighting factor, sensor error weighting factor, and communication quality weighting factor, respectively; i represents the number of the dynamic virtual line-of-sight segment, i = 1, 2, ..., j, where j is a positive integer greater than 2; a represents the number of the vehicle in the dynamic virtual line-of-sight segment, a = 1, 2, ..., b, where b is a positive integer greater than 2; e represents a natural constant.

6. The vehicle collision warning method based on driving sight distance according to claim 1, characterized in that, In step 3, the traffic target data collected by the agent node within its dynamic virtual line-of-sight segment includes dynamic target type and quantity, key motion parameters, and abnormal event flags. The agent node then compresses the dynamic target type and quantity, key motion parameters, and abnormal event flags into summary information and transmits it to the edge node.

7. The vehicle collision warning method based on driving sight distance according to claim 6, characterized in that, In step 3, the process of the edge nodes constructing a global dynamic risk map is as follows: Step 3.1) Data Collection and Preprocessing Edge nodes periodically receive summary information uploaded by proxy nodes in each dynamic virtual line-of-sight segment, and perform time synchronization and spatial registration of the summary information to ensure that all data is mapped to a unified map road coordinate system and time axis. Step 3.2) Spatial Mapping and Partitioning Edge nodes spatially partition roads on the map according to dynamic virtual line-of-sight segments, so that each dynamic virtual line-of-sight segment partition corresponds to a different spatial unit on the map. Step 3.3) Risk Factor Calculation For each dynamic virtual line-of-sight segment, the edge nodes calculate the risk factor of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on the summary information; Step 3.4) Map fusion and dynamic updating Edge nodes map the risk factors of all dynamic virtual line-of-sight segments onto the global road map, forming a one-dimensional or two-dimensional global dynamic risk map.

8. The vehicle collision warning method based on driving sight distance according to claim 1, characterized in that, In step 4, the edge node predicts the future trajectory of all dynamic targets in each dynamic virtual line-of-sight segment; when the predicted future trajectory is linear, the edge node uses a machine learning model to predict the collision probability of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on the risk factor; when the predicted future trajectory is non-linear, the edge node uses an LSTM deep learning model to output the collision probability of the target vehicle in the corresponding dynamic virtual line-of-sight segment based on the risk factor. Finally, by combining the environmental credibility factor of the edge nodes, the predicted collision probability of the target vehicle in each dynamic virtual line-of-sight segment is calculated.

9. The vehicle collision warning method based on driving sight distance according to claim 1, characterized in that, In step 4, the edge node compares the predicted collision probability of the target vehicle in each dynamic virtual line-of-sight segment with the preset predicted collision probability ranges for non-warning level, prompt level, warning level, and emergency level, respectively. If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset non-warning level predicted collision probability range, then the warning level of the target vehicle in that dynamic virtual line-of-sight segment is determined to be non-warning level. If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset warning level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the warning level. If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset warning level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the warning level. If the predicted collision probability of the target vehicle in a certain dynamic virtual line-of-sight segment is within the preset emergency level predicted collision probability range, then the warning for the target vehicle in that dynamic virtual line-of-sight segment is determined to be at the emergency level.

Images