Vehicle behavior recognition method and system based on intelligent expressway system

Abstract

The present application relates to the field of traffic control, and particularly relates to a vehicle behavior recognition method and system based on an intelligent expressway system, the method comprising: collecting state information of all vehicles located on a set road section at a fixed frequency, the state information comprising a heading angle and a planar coordinate and a horizontal velocity vector of the vehicle in a planar rectangular coordinate system with the starting point of the set road section as the origin; calculating a real-time lane-changing probability of a target vehicle, and determining that the target vehicle has a high lane-changing risk when the real-time lane-changing probability is greater than or equal to a preset threshold value. The present application does not rely on fixed road markings such as physical lane lines at all, and only performs traffic flow modeling and behavior analysis based on the motion trajectory of the vehicle itself, has excellent all-weather and all-road section adaptability, and can timely identify lane-changing risks by calculating the lane-changing probability of the vehicle in real time, providing a longer reaction time for the traffic management system and the rear vehicle, and effectively improving road safety.

Description

Technical Field

[0001] This invention relates to the field of traffic control technology. In particular, it relates to a vehicle behavior recognition method and system based on an intelligent highway system. Background Technology

[0002] With the development of intelligent transportation systems, highway vehicle behavior recognition technology has gradually become a key means to improve road safety and traffic efficiency. Traditional vehicle behavior recognition methods mostly rely on sensors such as fixed lane lines, cameras, or radar to issue alerts after the fact for obvious behaviors such as vehicles crossing lane lines or deviating from lanes.

[0003] Chinese patent application CN116189444A discloses an automatic vehicle dynamic information identification method, which includes: performing vehicle attitude detection and trajectory tracking monitoring to obtain the vehicle's position; collecting vehicle motion information, including speed and acceleration, in the lane environment where the vehicle is located; the vehicle monitoring system collecting the dynamic vehicle's driving information and uploading the collected information after preliminary judgment; judging abnormal driving behavior of the vehicle based on the vehicle's position information, motion information, and lane environment, predicting the vehicle's motion state, calculating the Mahalanobis distance between the predicted position and the detected position; if the Mahalanobis distance is less than a set threshold, the vehicle's motion state is successfully associated, and the predicted position is taken as the vehicle's position; and using an instance segmentation network model to analyze and judge the vehicle's position information, motion information, and lane environment to obtain abnormal driving behavior of the vehicle; abnormal driving behavior includes abnormal lane changes and stopping.

[0004] However, existing recognition methods mostly rely on fixed road infrastructure, such as lane markings, cameras, or radar sensors. These methods generally suffer from recognition lag, typically triggering warnings only after a vehicle has already made a significant lane departure or crossed the line. This makes it difficult to provide sufficient reaction time for traffic management systems and vehicles behind. Furthermore, the recognition performance of traditional methods deteriorates significantly under low visibility conditions such as unclear road markings, inclement weather, or at night, resulting in poor adaptability. Summary of the Invention

[0005] To address the technical problem of poor vehicle behavior recognition performance of the aforementioned recognition methods, the present invention provides solutions in the following aspects.

[0006] In the first aspect, the vehicle behavior recognition method based on the intelligent highway system includes:

[0007] The status information of all vehicles located on a designated road segment is collected at a fixed frequency. The status information includes the heading angle, as well as the plane coordinates and horizontal velocity vector of the vehicle in a Cartesian coordinate system with the origin of the designated road segment as the origin.

[0008] Calculate the real-time lane change probability of the target vehicle. When the real-time lane change probability is greater than or equal to a preset threshold, the target vehicle is determined to have a high lane change risk.

[0009] The method for calculating the real-time lane change probability of the target vehicle includes: for each vehicle, obtaining short-time motion segments that meet the criteria for uniform linear motion based on the vehicle's heading angle and horizontal velocity vector, and marking them as uniform motion segments; calculating the BeiDou error of a single vehicle based on all uniform motion segments; obtaining vehicles belonging to uniform motion segments at the current moment, and using a clustering algorithm to divide these vehicles into different traffic flow clusters based on the vehicle's current planar coordinates and horizontal velocity vector; for each traffic flow cluster, using the current planar coordinates of vehicles within the cluster and the BeiDou error to fit and generate the current traffic flow curve, and generating traffic flow dividing lines based on adjacent traffic flow curves; obtaining the planar coordinates and heading angle of the target vehicle within a preset time window with the current moment as the endpoint, and combining the current traffic flow curve and traffic flow dividing lines to calculate the heading angle change rate and lateral distance change rate, and fusing the heading angle change rate and lateral distance change rate to obtain the real-time lane change probability.

[0010] Preferably, the method for obtaining uniform motion segments includes: for a single vehicle, setting a short-time sliding window, obtaining the heading angle time sequence and speed time sequence of the vehicle on a set road segment up to the current time, sliding the short-time sliding window with the time interval between adjacent acquisition times as the step size starting from the beginning of the heading angle time sequence and speed time sequence, to obtain multiple short-time motion segments; calculating the uniform straightness of the short-time motion segments, and marking the short-time motion segments whose uniform straightness meets the uniform linear motion judgment condition as uniform motion segments.

[0011] Preferably, the method for calculating uniform velocity straightness includes: calculating the average value of all heading angles within a short-time motion segment; calculating the absolute deviation of each heading angle within the short-time motion segment from the aforementioned average value; calculating the average value of all absolute deviations and performing positive correlation normalization to obtain a normalized value; acquiring the velocity at all acquisition times within the short-time motion segment; for each pair of adjacent velocities, calculating the ratio of the velocity at the later acquisition time to the velocity at the previous acquisition time; calculating the absolute difference between this ratio and 1; calculating the average value of all absolute differences; and using the product of this average value and the normalized value as the uniform velocity straightness.

[0012] Preferably, the method for calculating the BeiDou error of a vehicle includes: obtaining the planar coordinates of the starting points of all uniform motion segments of a single vehicle; constructing corresponding ideal uniform motion segments with the starting point of each uniform motion segment as the starting point; dividing the ideal uniform motion segments into a preset number of equal intervals; calculating the ideal planar coordinates of the interval points between the starting point and the ending point based on the planar coordinates of the starting point and the horizontal velocity vector; calculating the average Euclidean distance between the ideal planar coordinates of all interval points and the planar coordinates at the corresponding acquisition time to obtain the BeiDou error of the vehicle.

[0013] Preferably, the method for dividing traffic flow clusters includes: calculating the dominant velocity direction unit vector based on the horizontal velocity vector of all vehicles moving at a constant linear speed at the current moment; constructing a lateral unit vector orthogonal to the dominant velocity direction unit vector; projecting the current planar coordinates of the vehicles onto the lateral unit vector direction to obtain the lateral projection value of the vehicles; using the lateral projection values ​​of all vehicles as a sample set; clustering the sample set using the mean-shift clustering algorithm; and assigning the vehicles to the corresponding traffic flow clusters based on the clustering results.

[0014] Preferably, the fitting method for the current traffic flow curve includes: fitting the planar coordinates of all vehicles in a single traffic flow cluster at the current time to obtain a fitting curve; calculating and summing the vertical Euclidean distances from the planar coordinates of all vehicles in the single traffic flow cluster at the current time to the fitting curve to obtain the overall loss; removing the planar coordinates of outliers based on the vertical Euclidean distance, fitting the remaining planar coordinates to obtain a new fitting curve and calculating the overall loss; iterating this step until the difference in the ratio of the overall loss of two adjacent iterations in the most recent three consecutive iterations is less than a preset threshold; and using the fitting curve obtained in the last iteration as the traffic flow curve.

[0015] Preferably, the method for calculating the heading angle change rate includes: calculating the first absolute deviation angle between the heading angle at the end of the preset time window and the tangent direction angle of the end point in the traffic flow curve to which the vehicle belongs; calculating the second absolute deviation angle between the heading angle at the start of the preset time window and the tangent direction angle of the start point in the traffic flow curve to which the vehicle belongs; calculating the difference between the first absolute deviation angle and the second absolute deviation angle; and using the ratio of the difference to the second absolute deviation angle as the heading angle change rate.

[0016] Preferably, the method for calculating the lateral distance change rate includes: obtaining the planar coordinates of the start and end points of a preset time window in the traffic flow curve to which the vehicle belongs; calculating the first vertical Euclidean distance between the end point and the nearest traffic flow dividing line; calculating the second vertical Euclidean distance between the start point and the nearest traffic flow dividing line; calculating the difference between the first and second vertical Euclidean distances; and using the ratio of the difference to the second vertical Euclidean distance as the lateral distance change rate.

[0017] Preferably, the step of fusing the rate of change of heading angle and the rate of change of lateral distance to obtain the real-time lane change probability includes: calculating the product of the rate of change of heading angle and the rate of change of lateral distance, and performing positive correlation normalization on the absolute value of the product to obtain the real-time lane change probability.

[0018] Secondly, a vehicle behavior recognition system based on an intelligent highway system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned vehicle behavior recognition method based on the intelligent highway system is implemented.

[0019] The present invention has the following effects:

[0020] 1. This invention does not rely on physical lane lines or other fixed road markings at all. It performs traffic flow modeling and behavior analysis solely based on the vehicle's own movement trajectory, enabling the system to work reliably in low visibility weather such as night, rain, snow, and fog, as well as in complex scenarios such as worn or missing road markings and curves. It has excellent all-weather and all-road adaptability.

[0021] 2. By calculating the lane change probability of vehicles in real time, this invention can identify lane change risks even before a vehicle crosses the lane line and only shows the intention to change lanes, providing a longer reaction time for the traffic management system and vehicles behind, and effectively improving road safety.

[0022] 3. This invention comprehensively judges lane change behavior by integrating the rate of change of heading angle and the rate of change of lateral distance, and performs dynamic calibration by combining the BeiDou error of individual vehicles. This effectively reduces the possibility of misjudging normal driving swaying or driving on curves within the lane as lane change, and significantly improves the accuracy and reliability of behavior recognition. Attached Figure Description

[0023] Figure 1 This is a flowchart of steps S1-S2 in the vehicle behavior recognition method based on a smart highway system according to an embodiment of the present invention.

[0024] Figure 2 This is a schematic diagram of steps S20-S24 in the vehicle behavior recognition method based on a smart highway system according to an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] Reference Figure 1The vehicle behavior recognition method based on the intelligent highway system includes steps S1-S2, as follows:

[0028] S1: Collect the status information of all vehicles located on the set road segment at a fixed frequency. The status information includes the heading angle, the plane coordinates of the vehicle in the Cartesian coordinate system with the starting point of the set road segment as the origin, and the horizontal velocity vector.

[0029] Vehicle movement on the road is continuous and dynamic, especially during critical maneuvers such as lane changes, rapid acceleration, or emergency braking, where their state can change significantly within a hundredth of a second. Therefore, using the BeiDou system... The system collects status information of all vehicles passing through the designated road section at a high frequency. This ultra-high sampling rate can accurately record the complete trajectory and status changes of vehicle movement, ensuring that every subtle driving operation is fully captured.

[0030] The status information includes the heading angle and the vehicle's planar coordinates and horizontal velocity vector in a Cartesian coordinate system with the origin at the starting point of the designated road segment. Since the vehicle's position and velocity information collected by the BeiDou system is in spherical coordinates, it cannot be directly used for subsequent calculations. Therefore, the position and velocity information needs to be converted to the local ENU coordinate system (East-North-Up). The local ENU coordinate system has the origin at the starting point of the designated road segment, with the E-axis pointing east, the N-axis pointing north, and the U-axis pointing vertically upwards. After conversion to the local ENU coordinate system, the U-axis is ignored, and then the system is further converted to a Cartesian coordinate system with the E-axis as the x-axis and the N-axis as the y-axis. This yields the vehicle's planar coordinates and horizontal velocity vector in the Cartesian coordinate system, which are then used in subsequent calculations. The velocity and velocity direction can be calculated from the horizontal velocity vector; the calculation process is existing technology and will not be elaborated here.

[0031] S2: Calculate the real-time lane change probability of the target vehicle. When the real-time lane change probability is greater than or equal to a preset threshold, the target vehicle is determined to have a high lane change risk.

[0032] Traditional systems only issue warnings when a vehicle has already crossed the lane line or collided with another vehicle. This invention, by calculating the real-time lane-changing probability of the target vehicle, identifies lane-changing risks when the vehicle intends to change lanes but has not yet crossed the lane line. This allows valuable reaction time for following vehicles or the system, providing valuable data for traffic theory research.

[0033] See Figure 2 Step S2 includes steps S20-S24, as detailed below:

[0034] S20: For each vehicle, based on the vehicle's heading angle and horizontal velocity vector, obtain short-time motion segments that meet the criteria for uniform linear motion and mark them as uniform motion segments.

[0035] The method for obtaining uniform motion segments includes: for a single vehicle, setting a short-term sliding window (e.g., a 1-second sliding window), obtaining the heading angle time series and speed time series of the vehicle on a set road segment up to the current time, sliding the short-term sliding window with the time interval between adjacent acquisition times as the step size starting from the beginning of the heading angle time series and speed time series, to obtain multiple short-term motion segments; calculating the uniform straightness of each short-term motion segment based on the heading angle and speed within the short-term motion segment; when the uniform straightness meets the uniform linear motion judgment condition, marking the corresponding short-term motion segment as a uniform motion segment.

[0036] The method for calculating the uniform straightness of a short-term motion segment includes: calculating the average of all heading angles within the short-term motion segment; calculating the absolute deviation of each heading angle from the aforementioned average value within the short-term motion segment; calculating the ratio of the average of all absolute deviations to the steering angle corresponding to the vehicle's maximum turning radius, obtaining a first ratio; calculating the velocity at each acquisition moment based on the horizontal velocity vector within the short-term motion segment; for every two adjacent velocities within the short-term motion segment, calculating the ratio of the velocity at the later acquisition moment to the velocity at the previous acquisition moment, obtaining a second ratio; calculating the absolute difference between the second ratio and 1; calculating the average of all absolute differences; and using the product of this average value and the first ratio as the uniform straightness. The specific formula is as follows:

[0037]

[0038] In the formula, Indicates the uniform straightness of a short-term motion segment; Indicates the first time segment of a short motion segment The heading angle at each data acquisition moment; It represents the average value of all heading angles within a short motion segment; Indicates the first The absolute deviation of the heading angle at each acquisition time from the average of all heading angles; This represents the average of all absolute deviation values ​​within a short-term motion segment, reflecting the average fluctuation of the vehicle's heading angle within this short-term motion segment. This represents the steering angle corresponding to the vehicle's maximum turning radius. Since the maximum change in steering angle within a short period cannot exceed the steering angle corresponding to the vehicle's maximum permissible turning radius, it is used... Normalize the average fluctuation range of the heading angle; Indicates the first time segment of a short motion sequence The speed at each acquisition moment; Indicates the first time segment of a short motion sequence The speed at each data collection moment; as the vehicle approaches uniform linear motion, The closer it is to 1, the better. To measure the rate of change of velocity between adjacent acquisition times; It represents the average of the absolute difference between the ratio of each two adjacent velocities within a short motion segment and 1, reflecting the overall rate of change of vehicle speed within that short motion segment.

[0039] By dividing by the steering angle corresponding to the vehicle's maximum turning radius, adaptation to different vehicle types is achieved. When trucks and cars have the same average yaw angle fluctuation, the average fluctuation may be close to the operating limit because the steering angle corresponding to the truck's maximum turning radius is relatively small, while the car is still within the controllable range. After normalizing the average fluctuation by using the steering angle corresponding to each vehicle's own maximum turning radius, the truck's score may be higher, making it easier to identify as a non-uniform speed state, which is consistent with the understanding of the dynamic characteristics of different vehicles.

[0040] Of course, it is understandable that in other embodiments, the average fluctuation of the heading angle can also be normalized by dividing by the standard deviation of all heading angles within the short-term running segment.

[0041] The criterion for determining uniform linear motion is that the uniform linearity of a short-time motion segment is less than or equal to a threshold of 0.2. Therefore, the uniform linearity of a short-time motion segment... When the vehicle is determined to be moving at a constant velocity in a straight line within that short motion segment, the short motion segment is marked as a uniform motion segment. If the vehicle is not moving at a constant speed in a straight line within that short motion segment, it is determined that the vehicle is not moving at a constant speed in a straight line.

[0042] This threshold allows for minor speed fluctuations and heading adjustments in a short period (for example, a vehicle traveling smoothly on a high-speed curve may have a constantly changing heading angle, but its linearity within a 1-second time window is very high, and it can still be considered to be in a state of uniform linear motion). This design ensures that the method of this invention can stably acquire effective data in various highway scenarios, including straight roads and curves, and has strong environmental adaptability.

[0043] It should be noted that the uniform linear motion described in this invention is not an absolute ideal physical state, but rather a stable motion state in which the vehicle's motion state (speed and heading) does not change significantly under short-term, high-frequency observation scales. This definition aims to extract high-quality motion segments that can be used for high-precision modeling from continuous dynamic vehicle trajectories.

[0044] This invention achieves high-precision and robust identification of uniform linear motion of a vehicle by fusing the average fluctuation range of the normalized vehicle heading angle with the overall rate of change of vehicle speed. The vehicle is only judged to be in uniform linear motion when it simultaneously meets the requirements of directional stability and speed stability, effectively avoiding misjudgment caused by a single indicator.

[0045] S21: Calculate the BeiDou error of a single vehicle based on all uniform motion segments of that vehicle.

[0046] Based on the uniform motion segments of each vehicle obtained in step S20, the BeiDou error of each vehicle is calculated. Specifically: the planar coordinates of the starting points of all uniform motion segments of a single vehicle are obtained; using the planar coordinates of the starting points of each uniform motion segment as the starting point, an ideal uniform motion segment with the same time length as the uniform motion segment is constructed; the time length of the ideal uniform motion segment is divided into a preset number of parts (e.g., 5 parts) at equal intervals; the ideal planar coordinates of the interval points between the starting and ending points are calculated based on the planar coordinates of the starting points and the horizontal velocity vector; the planar coordinates of the acquisition time corresponding to the interval points in the uniform motion segment are obtained; the average Euclidean distance between the ideal planar coordinates of all interval points and the planar coordinates of the corresponding acquisition time is calculated to obtain the BeiDou error of the vehicle. The specific formula is as follows:

[0047]

[0048] In the formula, Indicates the first BeiDou error of individual vehicles; Indicates the first The total number of uniform motion segments of a vehicle (i.e., the total number of ideal uniform motion segments). This indicates that there are 4 interval points in the ideal uniform motion segment corresponding to each uniform motion segment; This represents the total number of intervals in all ideal uniform motion segments; Indicates the first The Euclidean distance between the ideal planar coordinates of each interval point and the planar coordinates at the corresponding acquisition time.

[0049] Since we assume that the vehicle moves at a constant speed along a straight line in the uniform motion segment, the calculation of the ideal plane coordinates of the interval point between the starting point and the ending point based on the plane coordinates of the starting point and the horizontal velocity vector includes: obtaining the time from a single interval point to the starting point, as well as the velocity and velocity direction of the starting point; multiplying the velocity of the starting point by the time to obtain the ideal displacement of the vehicle from the starting point to the interval point; and shifting the plane coordinates of the starting point along the velocity direction of the starting point by the corresponding ideal displacement to obtain the ideal plane coordinates of the interval point.

[0050] Dynamic calibration of individual BeiDou errors is achieved by constructing an ideal trajectory for each vehicle based on a uniform motion segment and calculating its average deviation from the measured BeiDou trajectory. This method fully utilizes the data reliability and motion predictability of high-quality motion segments, resulting in BeiDou errors with high individual correlation, high accuracy, and strong noise resistance. It provides a crucial high-quality data foundation and adaptive threshold parameters for accurate fitting of subsequent traffic flow curves, thereby improving the overall accuracy and robustness of the vehicle behavior recognition system.

[0051] S22: Obtain the vehicles that belong to the uniform motion segment at the current moment, and use a clustering algorithm to divide these vehicles into different traffic clusters based on the vehicle's current planar coordinates and horizontal velocity vector.

[0052] For all vehicles moving at a constant velocity in a straight line at the current moment, calculate the average of the horizontal velocity vectors of all vehicles to obtain the dominant velocity direction unit vector. Construct a lateral unit vector orthogonal to the dominant velocity direction unit vector, and project the vehicle's current planar coordinates onto the lateral unit vector direction to obtain the vehicle's lateral projection value. Using the lateral projection values ​​of all vehicles moving at a constant velocity in a straight line at the current moment as a sample set, cluster the sample set using the mean-drift clustering algorithm. Calculate the bandwidth using Silverman's rule of thumb and estimate the density using a Gaussian kernel function to drive the mean-drift clustering process. Iteratively calculate the density gradient of each sample and move it towards its local density maxima, ultimately causing all samples to converge to several stable density extrema points. The number of these extrema points directly represents the effective traffic flow number. Based on the vehicle's lateral offset, assign the vehicle to the traffic flow cluster closest to the extrema point, completing the precise matching of vehicles and traffic flows. This yields the current traffic flow cluster, the vehicles contained in the traffic flow cluster, and the vehicle's current planar coordinates.

[0053] Traditional solutions heavily rely on clear, visible lane markings. This method completely eliminates the reliance on any physical or electronic lane markings, relying solely on the vehicle's own trajectory. This allows the system to work reliably even at night, in low-visibility weather conditions such as rain, snow, and fog, when road markings are worn or missing, and in complex curve scenarios, greatly improving the system's all-weather and all-road applicability.

[0054] S23: For each traffic flow cluster, the current traffic flow curve is generated by fitting the current planar coordinates of the vehicles in the traffic flow cluster with the Beidou error, and a traffic flow dividing line is generated based on the adjacent traffic flow curves.

[0055] The process of generating the current traffic flow curve by fitting the current planar coordinates of vehicles within a traffic flow cluster and the BeiDou error involves the following steps: For a single traffic flow cluster, the planar coordinates of all vehicles within the cluster at the current moment are fitted using the least squares method to obtain a fitted curve. The vertical Euclidean distances from the current planar coordinates of all vehicles within the cluster to the fitted curve are calculated and summed to obtain the overall loss. The planar coordinates corresponding to the maximum vertical Euclidean distance and those with vertical Euclidean distances greater than the vehicle's BeiDou error are removed. The remaining planar coordinates are then fitted to obtain a new fitted curve, and the overall loss is calculated. This step is iterated until the difference in the ratio of the overall loss between two consecutive iterations in the most recent three iterations is less than a preset threshold. The fitted curve obtained in the last iteration is taken as the traffic flow curve. The specific formula is as follows:

[0056]

[0057] In the formula, Indicates the first Convergence stability after the next iteration; Indicates the first The overall loss of the next iteration; Indicates the first The overall loss of the next iteration; Indicates the first The overall loss of the next iteration; Measured from the first The iteration to the... The percentage reduction in overall loss in each iteration; Measured from the first The iteration to the... The percentage reduction in overall loss in each iteration.

[0058] The rate of change of convergence speed was measured when A larger value indicates a faster convergence speed, with each optimization step performing better than the previous one. The system is in a highly efficient learning phase, rapidly eliminating major outliers (coordinates with the largest vertical Euclidean distance or exceeding their BeiDou error). When the value is close to zero, it means that the convergence rate remains constant, indicating that the optimization process has entered a stable state. Therefore, in this embodiment, the preset threshold is set to 0.1. When the time is reached, stop iterating to obtain the final traffic flow curve.

[0059] After obtaining the traffic flow curves corresponding to all traffic flow clusters, coordinate points on a single traffic flow curve are collected at a preset density to obtain the coordinate point set of the traffic flow curve. For two adjacent traffic flow curves, the midpoint of the two coordinate points at the same collection time in the coordinate point set of the two traffic flow curves is calculated to obtain the midpoint sequence. The midpoints in the midpoint sequence are connected according to the collection order to obtain the traffic flow dividing line between two adjacent traffic flow curves.

[0060] The traffic flow dividing line is derived entirely from the real-time fitted traffic flow curve, accurately reflecting the actual spatial division between adjacent traffic flows, and providing a precise geometric reference for subsequent calculation of the distance from vehicles to the traffic flow dividing line.

[0061] S24: Obtain the planar coordinates and heading angle of the target vehicle within a preset time window with the current time as the endpoint, and combine the current traffic flow curve and traffic flow dividing line to calculate the heading angle change rate and lateral distance change rate, and fuse the heading angle change rate and lateral distance change rate to obtain the real-time lane change probability.

[0062] On highways, lane changing is a continuous action completed within a short period (typically 3-4 seconds). During this process, the vehicle's heading angle and its lateral distance from the traffic flow dividing line both undergo significant and coordinated changes. Based on this physical characteristic, for a target vehicle, its planar coordinates and heading angle are obtained within a preset time window (e.g., a 3-second time window) with the current moment as the endpoint.

[0063] The method for calculating the heading angle change rate includes: calculating the first absolute deviation angle between the heading angle at the end of the preset time window and the tangent direction angle of the end point in the traffic flow curve to which the target vehicle belongs; calculating the second absolute deviation angle between the heading angle at the start of the preset time window and the tangent direction angle of the start point in the traffic flow curve to which the target vehicle belongs; calculating the difference between the first absolute deviation angle and the second absolute deviation angle; and using the ratio of this difference to the second absolute deviation angle as the heading angle change rate.

[0064] The method for calculating the lateral distance change rate includes: obtaining the position points of the planar coordinates of the start and end points of the preset time window in the traffic flow curve to which the target vehicle belongs; calculating the first vertical Euclidean distance between the position point corresponding to the end point and the nearest traffic flow dividing line; calculating the second vertical Euclidean distance between the position point corresponding to the start point and the nearest traffic flow dividing line; calculating the difference between the first vertical Euclidean distance and the second vertical Euclidean distance; and using the ratio of the difference to the second vertical Euclidean distance as the lateral distance change rate.

[0065] The real-time lane change probability is obtained by fusing the rate of change of heading angle and the rate of change of lateral distance by calculating the product of the rate of change of heading angle and the rate of change of lateral distance, and then performing positive correlation normalization on the absolute value of the product to obtain the real-time lane change probability.

[0066] When the preset time window is 3 seconds long, the real-time lane change probability is calculated using the following formula:

[0067]

[0068] In the formula, This indicates the real-time probability of the target vehicle changing lanes. It represents the deviation angle between the heading angle of the 3rd second within the time window (i.e., the end of the preset time window) and the tangent angle of the position point of the target vehicle in the traffic flow curve at that time point; It represents the deviation angle between the heading angle of the first second within the time window (i.e., the start of the preset time window) and the tangent angle of the position point of the target vehicle in the traffic flow curve at that time point; It represents the vertical Euclidean distance between the position of the planar coordinates corresponding to the 3rd second within the time window (i.e., the end of the preset time window) in the traffic flow curve to which the target vehicle belongs and the nearest traffic flow dividing line; It represents the vertical Euclidean distance between the position of the planar coordinates corresponding to the 1st second within the time window (i.e., the start of the preset time window) in the traffic flow curve to which the target vehicle belongs and the nearest traffic flow dividing line. This represents the hyperbolic tangent function, used to map the obtained value to... Within the range.

[0069] It reflects the proportional change in the degree of deviation of the heading angle, eliminating the influence of the target vehicle's initial heading state (such as being traveling in a state of slight deviation) on the absolute change; It reflects the proportional change in the distance from the target vehicle to the traffic flow dividing line, eliminating the distance benchmark difference caused by the initial location in the center or edge of the lane, and fairly assessing the lateral movement trend.

[0070] This indicates that steering intention (change in heading angle) and lateral displacement (change in distance) must occur in tandem, which is consistent with the physical nature of lane change behavior. This significantly reduces the probability of misjudging slight lane swaying or cornering as lane change, and improves the system's specificity and robustness.

[0071] When the real-time lane change probability is greater than or equal to a preset threshold (e.g., 0.6), the target vehicle is determined to have a high lane change risk, i.e., a high lane change probability; when the real-time lane change probability is less than the preset threshold, the target vehicle is determined to have a low lane change risk, i.e., a low lane change probability.

[0072] By integrating the rate of change of heading angle and the rate of change of lateral distance to comprehensively judge lane change behavior, and combining the BeiDou error of individual vehicles for dynamic calibration, the possibility of misjudging normal driving swaying or driving on curves within the lane as lane change is effectively reduced, and the accuracy and reliability of behavior recognition are significantly improved.

[0073] This application also discloses a vehicle behavior recognition system based on an intelligent highway system. The system includes a processor and a memory. The memory stores computer program instructions. When the computer program instructions are executed by the processor, the vehicle behavior recognition method based on an intelligent highway system according to the above embodiments of the present invention is implemented.

[0074] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0075] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A vehicle behavior recognition method based on an intelligent highway system, characterized in that, include: The status information of all vehicles located on a set road segment is collected at a fixed frequency. The status information includes the heading angle, as well as the plane coordinates and horizontal velocity vector of the vehicle in a Cartesian coordinate system with the starting point of the set road segment as the origin. Calculate the real-time lane change probability of the target vehicle. When the real-time lane change probability is greater than or equal to a preset threshold, the target vehicle is determined to have a high lane change risk. The method for calculating the real-time lane change probability of the target vehicle includes: for each vehicle, obtaining short-time motion segments that meet the criteria for uniform linear motion based on the vehicle's heading angle and horizontal velocity vector, and marking them as uniform motion segments; calculating the BeiDou error of a single vehicle based on all uniform motion segments; obtaining vehicles belonging to uniform motion segments at the current moment, and using a clustering algorithm to divide these vehicles into different traffic flow clusters based on the vehicle's current planar coordinates and horizontal velocity vector; for each traffic flow cluster, using the current planar coordinates of vehicles within the cluster and the BeiDou error to fit and generate the current traffic flow curve, and generating traffic flow dividing lines based on adjacent traffic flow curves; obtaining the planar coordinates and heading angle of the target vehicle within a preset time window with the current moment as the endpoint, and combining the current traffic flow curve and traffic flow dividing lines to calculate the heading angle change rate and lateral distance change rate, and fusing the heading angle change rate and lateral distance change rate to obtain the real-time lane change probability.

2. The vehicle behavior recognition method based on a smart highway system according to claim 1, characterized in that, The method for obtaining uniform motion segments includes: for a single vehicle, setting a short-time sliding window, obtaining the heading angle time sequence and speed time sequence of the vehicle on a set road segment up to the current time, sliding the short-time sliding window with the time interval between adjacent acquisition times as the step size starting from the beginning of the heading angle time sequence and speed time sequence, to obtain multiple short-time motion segments; calculating the uniform straightness of the short-time motion segments, and marking the short-time motion segments whose uniform straightness meets the uniform linear motion judgment condition as uniform motion segments.

3. The vehicle behavior recognition method based on a smart highway system according to claim 2, characterized in that, The method for calculating uniform velocity straightness includes: calculating the average value of all heading angles within a short-time motion segment; calculating the absolute deviation of each heading angle within the short-time motion segment from the aforementioned average value; calculating the average value of all absolute deviations and performing positive correlation normalization to obtain the normalized value; acquiring the velocity at all acquisition times within the short-time motion segment; for each two adjacent velocities, calculating the ratio of the velocity at the later acquisition time to the velocity at the previous acquisition time; calculating the absolute difference between this ratio and 1; calculating the average value of all absolute differences; and using the product of this average value and the normalized value as the uniform velocity straightness.

4. The vehicle behavior recognition method based on a smart highway system according to claim 1, characterized in that, The method for calculating the BeiDou error of a vehicle includes: obtaining the planar coordinates of the starting points of all uniform motion segments of a single vehicle; constructing corresponding ideal uniform motion segments with the starting point of each uniform motion segment as the starting point; dividing the ideal uniform motion segments into a preset number of equal intervals; calculating the ideal planar coordinates of the interval points between the starting point and the ending point based on the planar coordinates of the starting point and the horizontal velocity vector; and calculating the average Euclidean distance between the ideal planar coordinates of all interval points and the planar coordinates at the corresponding acquisition time to obtain the BeiDou error of the vehicle.

5. The vehicle behavior recognition method based on an intelligent highway system according to claim 1, characterized in that, The method for dividing traffic flow clusters includes: calculating the dominant velocity direction unit vector based on the horizontal velocity vector of all vehicles moving at a constant linear speed at the current moment; constructing a lateral unit vector orthogonal to the dominant velocity direction unit vector; projecting the current planar coordinates of the vehicles onto the lateral unit vector direction to obtain the lateral projection value of the vehicles; using the lateral projection values ​​of all vehicles as a sample set, clustering the sample set using the mean-shift clustering algorithm; and assigning vehicles to the corresponding traffic flow clusters based on the clustering results.

6. The vehicle behavior recognition method based on a smart highway system according to claim 1, characterized in that, The current traffic flow curve fitting method includes: fitting the planar coordinates of all vehicles in a single traffic flow cluster at the current time to obtain a fitted curve; calculating and summing the vertical Euclidean distances from the planar coordinates of all vehicles in the single traffic flow cluster at the current time to the fitted curve to obtain the overall loss; removing the planar coordinates of outliers based on the vertical Euclidean distance, fitting the remaining planar coordinates to obtain a new fitted curve and calculating the overall loss; iterating this step until the difference in the ratio of the overall loss of two adjacent iterations in the most recent three consecutive iterations is less than a preset threshold; and using the fitted curve obtained in the last iteration as the traffic flow curve.

7. The vehicle behavior recognition method based on a smart highway system according to claim 1, characterized in that, The method for calculating the heading angle change rate includes: calculating the first absolute deviation angle between the heading angle at the end of the preset time window and the tangent direction angle of the end point in the traffic flow curve to which the vehicle belongs; calculating the second absolute deviation angle between the heading angle at the start of the preset time window and the tangent direction angle of the start point in the traffic flow curve to which the vehicle belongs; calculating the difference between the first absolute deviation angle and the second absolute deviation angle; and using the ratio of this difference to the second absolute deviation angle as the heading angle change rate.

8. The vehicle behavior recognition method based on a smart highway system according to claim 7, characterized in that, The method for calculating the lateral distance change rate includes: obtaining the planar coordinates of the start and end points of a preset time window in the traffic flow curve to which the vehicle belongs; calculating the first vertical Euclidean distance between the end point and the nearest traffic flow dividing line; calculating the second vertical Euclidean distance between the start point and the nearest traffic flow dividing line; calculating the difference between the first and second vertical Euclidean distances; and using the ratio of this difference to the second vertical Euclidean distance as the lateral distance change rate.

9. The vehicle behavior recognition method based on a smart highway system according to claim 8, characterized in that, The process of fusing the rate of change of heading angle and the rate of change of lateral distance to obtain the real-time lane change probability includes: calculating the product of the rate of change of heading angle and the rate of change of lateral distance, and performing positive correlation normalization on the absolute value of the product to obtain the real-time lane change probability.

10. A vehicle behavior recognition system based on an intelligent highway system, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the vehicle behavior recognition method based on a smart highway system according to any one of claims 1-9.

Images