Traffic monitoring method and system for vehicle tracking and early warning

Abstract

The invention relates to the technical field of traffic monitoring, in particular to a traffic monitoring method and system for vehicle tracking and early warning, and the method comprises the following steps: obtaining the real-time position, speed and acceleration data of all vehicles in a monitored road area, and setting according to the spatial distance between the vehicles and a time window; adjacent vehicles are aggregated into a dynamic traffic microscopic cluster, and a regional motion state reference is established; according to the method, the spatial distance and the time window between the vehicles are set, the dynamic traffic microscopic clusters are divided, refinement and dynamics of vehicle clustering are achieved, and the sensitivity and the response speed of road microscopic traffic environment changes are improved; and deviation calculation and weighted summation operation are carried out on a single vehicle motion vector by using a regional motion state reference, so that accurate quantification of individual driving behavior instability is realized, the abnormal driving state of the vehicle is captured, and the abnormal behavior recognition accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic monitoring, and in particular to a traffic monitoring method and system for vehicle tracking and early warning. Background Art

[0002] The field of traffic monitoring technology involves real-time or non-real-time monitoring, data collection, status analysis, and event identification of vehicles, pedestrians, and road infrastructure in the road traffic environment. It aims to improve road traffic safety, optimize traffic flow efficiency, and reduce the risk of traffic accidents. It is an important part of smart transportation.

[0003] Existing technologies are limited to simple collection and general analysis of vehicle location, traffic volume, and speed. This results in a single-dimensional approach to traffic status analysis, leading to delayed or inaccurate warnings of abnormal driving behavior. For example, it is difficult to quickly identify and locate abnormal driving behaviors such as sudden deceleration or drastic lane changes. This often leads to missed or false alarms, increasing the likelihood of traffic accidents. Therefore, improvements are needed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a traffic monitoring method and system for vehicle tracking and early warning.

[0005] To achieve the above object, the present invention adopts the following technical solution: a traffic monitoring method for vehicle tracking and early warning, comprising the following steps: Obtain real-time position, speed, and acceleration data for all vehicles within the monitored road area. Based on the spatial distance between vehicles and the time window setting, adjacent vehicles are aggregated into dynamic traffic micro-clusters to establish a regional motion state benchmark. Based on the regional motion state benchmark, a motion vector of a single vehicle is extracted to calculate an individual driving deviation value, and then the individual driving deviation values ​​of the single vehicle are summed up over a set time period to obtain a vehicle behavior instability measure; Based on the vehicle behavior instability measure of each vehicle, the vehicle behavior instability measures of all vehicles on the target road section are summed to obtain a cumulative driving risk of the road section, and the cumulative driving risk of the road section is then calculated with the real-time traffic volume, average speed and road capacity of the road section to generate a traffic flow conflict risk index for the road section; Based on the traffic flow conflict risk index of the road section, the traffic flow conflict risk index is compared one by one with the preset multi-level risk threshold value, and according to the comparison result, a preset control strategy set is matched and selected from the associated control strategy library. According to the preset control strategy set, adaptive traffic control instructions are generated for the traffic lights at the downstream intersection or the variable information signs ahead.

[0006] Preferably, the steps of obtaining the regional motion state benchmark are: Through the video sensors and millimeter-wave radar deployed in the road area, all vehicles currently traveling in the road area are detected synchronously and in real time, and the real-time position data, speed data and acceleration data of each vehicle are obtained. The data of each vehicle is uniformly marked with a timestamp to generate a set of real-time vehicle motion parameters. Based on the set of real-time vehicle motion parameters, according to a preset inter-vehicle spatial distance threshold and a time window threshold for vehicle motion state collection, vehicles whose spatial distance is below the spatial distance threshold and whose data collection time difference is within the time window threshold are identified as adjacent vehicles, and the adjacent vehicles are aggregated into multiple vehicle groups to generate dynamic traffic micro-clusters; Based on the dynamic traffic micro-cluster, speed data and acceleration data of all vehicles in each dynamic traffic micro-cluster are extracted to obtain a regional motion state benchmark.

[0007] Preferably, the steps for obtaining the individual driving deviation value are: Based on the regional motion state benchmark, the velocity component and acceleration component corresponding to a single vehicle at each time step are extracted, a four-dimensional motion vector of the vehicle at each time step is constructed, and the four-dimensional motion vectors corresponding to all time steps are integrated to form a continuous vehicle motion vector sequence to obtain a single vehicle motion vector sequence; According to the single vehicle motion vector sequence, the Mahalanobis distance between the vehicle motion vector at each time step and the average motion vector of the dynamic traffic micro-cluster to which it belongs is calculated to form an individual driving deviation value.

[0008] Preferably, the steps for obtaining the vehicle behavior instability measure are: Based on the individual driving deviation values, a vehicle behavior instability metric is calculated.

[0009] Preferably, the steps for obtaining the cumulative amount of road section driving risk are: Based on the vehicle behavior instability measurement of all vehicles in the target road section, the vehicle behavior instability measurement values ​​of all vehicles in the current time step are extracted, and the extracted values ​​are summed up to obtain the cumulative driving risk of the current road section in the current time step.

[0010] Preferably, the steps for obtaining the traffic flow conflict risk index are: According to the cumulative driving risk of the road section, the real-time traffic volume, road capacity and average speed of the road section at the corresponding time step are retrieved and included in the calculation input set together with the cumulative driving risk of the road section to obtain a joint variable set constituting the risk index calculation; Based on the joint variable set, a traffic flow conflict risk index is calculated.

[0011] Preferably, the step of obtaining the preset control strategy set is: Based on the traffic flow conflict risk index of the road section, a pre-defined multi-level risk threshold set is called, each risk threshold in the risk threshold set is extracted one by one, the traffic flow conflict risk index is compared with each risk threshold in terms of numerical value size respectively, the risk level matched with the current traffic flow conflict risk index is determined, and a road section risk level determination result is generated; According to the road section risk level determination result, the risk level determination result is taken as a query condition, a control strategy set corresponding to the risk level is called from an associated control strategy library, and a preset control strategy set suitable for the current road section risk level is formed.

[0012] Preferably, the step of obtaining the adaptive traffic control instruction is: Based on the preset control strategy set, strategy instructions in the control strategy set are extracted one by one, are respectively converted into instruction data streams that can be executed by a traffic control device, control actions on traffic signal lights of downstream intersections or variable information signs in front are generated one by one according to the instruction data streams, and an adaptive traffic control instruction is formed.

[0013] The application also provides a traffic monitoring system, comprising: A data acquisition module acquires real-time position, speed and acceleration data of all vehicles in a monitored road area, aggregates adjacent vehicles into dynamic traffic micro-clusters according to spatial distances between the vehicles and time window settings, and establishes a regional motion state reference; A behavior evaluation module extracts a motion vector of a single vehicle based on the regional motion state reference, calculates an individual driving deviation value, and sums continuous individual driving deviation values of the single vehicle within a set time period to obtain a vehicle behavior instability measure; A risk calculation module sums vehicle behavior instability measures of all vehicles on a target road section based on the vehicle behavior instability measures of the vehicles, obtains a road section driving risk accumulation, and calculates the road section driving risk accumulation with real-time traffic volume, average speed and road capacity of the road section to generate a traffic flow conflict risk index of the road section; A control instruction module compares the traffic flow conflict risk index of the road section with pre-set multi-level risk thresholds one by one, matches and selects a preset control strategy set from an associated control strategy library according to a comparison result, and generates an adaptive traffic control instruction on traffic signal lights of downstream intersections or variable information signs in front according to the preset control strategy set.

[0014] Compared with the prior art, the application has the advantages and positive effects that: The present invention divides dynamic traffic micro-clusters by setting spatial distances and time windows between vehicles, thereby achieving refinement and dynamism in vehicle clustering and improving sensitivity and response speed to changes in the road micro-traffic environment; using regional motion state benchmarks to calculate the deviation and perform weighted sum operations on the motion vector of a single vehicle, the present invention achieves accurate quantification of individual driving behavior instability, captures abnormal vehicle driving states, and improves the accuracy of abnormal behavior identification; at the road section level, the present invention integrates vehicle behavior instability with multi-dimensional traffic parameters such as real-time traffic volume, road capacity, and average speed, and generates a traffic flow conflict risk index in real time through collaborative computing, so that risk judgment is more in line with actual traffic conditions and the real-time and reliability of traffic risk warnings are improved; further, the traffic flow conflict risk index is compared step by step through multi-level risk thresholds, and control strategies are matched to dynamically generate control instructions for traffic lights at downstream intersections or variable information signs ahead, thereby achieving active and adaptive traffic intervention, reducing the probability of traffic accidents, and improving overall road traffic safety and traffic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the steps of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] See also Figure 1 The present invention provides a technical solution, a traffic monitoring method for vehicle tracking and early warning, comprising the following steps: Obtain real-time position, speed, and acceleration data for all vehicles within the monitored road area. Based on the spatial distance between vehicles and the time window setting, adjacent vehicles are aggregated into dynamic traffic micro-clusters to establish a regional motion state benchmark. Based on the regional motion state benchmark, the motion vector of a single vehicle is extracted and the individual driving deviation value is calculated. The continuous individual driving deviation values ​​of a single vehicle within a set time period are then summed to obtain a vehicle behavior instability measure. Based on the vehicle behavior instability measure of each vehicle, the vehicle behavior instability measures of all vehicles on the target road section are summed to obtain the cumulative driving risk of the road section. The cumulative driving risk of the road section is then calculated with the real-time traffic volume, average speed and road capacity of the road section to generate the traffic flow conflict risk index of the road section. Based on the traffic flow conflict risk index of the road section, the traffic flow conflict risk index is compared with the preset multi-level risk threshold one by one. According to the comparison result, the preset control strategy set is matched and selected from the associated control strategy library. According to the preset control strategy set, adaptive traffic control instructions are generated for the traffic lights at the downstream intersection or the variable information signs ahead.

[0018] The steps to obtain the regional motion state benchmark are: Through the video sensors and millimeter-wave radar deployed in the road area, all vehicles currently traveling in the road area are detected synchronously and in real time, and the real-time position data, speed data and acceleration data of each vehicle are obtained. The data of each vehicle is uniformly marked with a timestamp to generate a set of real-time vehicle motion parameters. Based on the real-time vehicle motion parameter set and the preset inter-vehicle spatial distance threshold and the time window threshold for vehicle motion state collection, vehicles whose spatial distance is lower than the spatial distance threshold and whose data collection time difference is within the time window threshold are identified as adjacent vehicles. Adjacent vehicles are aggregated into multiple vehicle groups to generate dynamic traffic micro-clusters. Based on dynamic traffic micro-clusters, the speed and acceleration data of all vehicles in each dynamic traffic micro-cluster are extracted to obtain the regional motion state benchmark.

[0019] Specifically, multiple sets of video sensors and millimeter-wave radars are deployed along the driving direction at key locations in the monitored road area, such as 100 to 300 meters upstream of major intersections, accident-prone sections, or highway ramp merge areas, to ensure that the sensor field of view fully covers all lanes and that there is at least 20% overlap between adjacent sensors. All sensors are connected to a unified network time protocol (NTP) server to synchronize the timestamps of the collected data, and the synchronization accuracy is controlled within 10 milliseconds. After the system is started, the video sensor continuously captures high-definition video streams of the road at a frequency of 25 frames per second. At the same time, the millimeter-wave radar transmits electromagnetic waves at a frequency of 20 Hz and receives target reflection signals. The system processes two data sources in parallel. For video stream data, a YOLOv5 target detection model that has been pre-trained using a large-scale traffic image dataset (such as BDD100K or KITTI dataset) is called to identify all vehicles in each frame of the image and generate a vehicle image for each vehicle. into a bounding box, then the DeepSORT tracking algorithm is used to predict the position of the vehicle in the next frame through Kalman filtering and combine the vehicle's appearance features to assign a unique tracking ID to each vehicle in a short time to achieve stable tracking across frames. Subsequently, the homography transformation matrix calculated in advance by the checkerboard calibration method is used to convert the central pixel coordinates of each vehicle's bounding box into world coordinates (x, y) on the actual road plane. By calculating the change in world coordinates between consecutive frames and dividing it by the frame interval time (40 milliseconds), the vehicle's velocity components on the x-axis and y-axis are obtained, and the velocity components are derived again to obtain the acceleration components. For millimeter-wave radar data, the system directly parses the target list it outputs to obtain the distance, azimuth, radial velocity and other information of each target, and uses this information to calculate the Cartesian coordinate position and velocity vector of the target vehicle. Then, an extended Kalman filter (EKF) is started for data fusion. The state vector of the filter is defined as The position measurements provided by the video data and the position and velocity measurements provided by the radar data are used as updated inputs to the EKF. Through fusion processing, a more accurate and robust set of motion state estimates is generated for each jointly detected vehicle. Finally, the unique tracking IDs, high-precision synchronized timestamps, fused position data, velocity data, and acceleration data of all vehicles at each time step (for example, every 50 milliseconds) are integrated together to form a set of real-time vehicle motion parameters.

[0020] Based on the set of real-time vehicle motion parameters generated in the previous process, the system starts the aggregation process of dynamic traffic micro-clusters. The core of this process is to judge the relevance of all vehicles based on two dynamically adjusted thresholds: the spatial distance threshold between vehicles and the time window threshold for vehicle motion status collection. First, the time window threshold for vehicle motion status collection is set. The setting of this threshold refers to the frequency of data collection and processing. For example, if the data update frequency of the real-time vehicle motion parameter set is 20 Hz (that is, one data point every 50 milliseconds), then the time window threshold is set to 150 milliseconds, that is, 3 data update cycles, to tolerate potential network transmission delays and calculation time, to ensure that the vehicle data used for comparison reflects the traffic conditions at similar times. Next, the system sets the spatial distance threshold between vehicles. This threshold is not a fixed value, but is related to the instantaneous speed of the vehicle. The specific calculation method is basic safety distance plus reaction distance, that is, spatial distance threshold = 5 meters + vehicle instantaneous speed × 1.5 seconds, where 5 meters is the minimum safe distance when the vehicle is stationary or traveling at low speed, and 1.5 seconds is the average reaction time of the driver during the following process. For example, if a car is traveling at 72 km / h (20 m / s), the corresponding spatial distance threshold is 5 + 20 × 1.5 = 35 meters. When the system aggregates, it takes any unclustered vehicle in the set as the core point, traverses all other vehicles in the vehicle real-time motion parameter set, and checks whether the absolute value of the difference between the data timestamps of each pair of vehicles is less than 150 milliseconds. If the time window requirement is met, the Euclidean distance between the two vehicles in the world coordinate system is further calculated, and this distance is compared with the dynamic spatial distance threshold calculated according to the core point vehicle speed. All vehicles that meet the threshold conditions in time and space are identified as adjacent vehicles, and all vehicles that can be reached through the adjacent relationship (that is, vehicles A and B are adjacent, B and C are adjacent, then A, B, and C belong to the same group) are classified into the same vehicle group. This process is repeated until all vehicles have been visited, and finally a dynamic traffic micro-cluster consisting of multiple independent vehicle groups is generated.

[0021] Based on the generated dynamic traffic micro-cluster, the system calculates the overall motion characteristics of each independent vehicle group to build a regional motion state benchmark. In specific execution, the system traverses each vehicle group in the dynamic traffic micro-cluster. For a selected group, it first extracts the speed data of these vehicles at the current time step (in the form of two-dimensional vectors) from the vehicle real-time motion parameter set based on the vehicle ID list contained in the group. form) and acceleration data (in the form of two-dimensional vector Before performing statistical calculations, an outlier removal step is performed to enhance the robustness of the benchmark. For speed data, the system calculates the median and interquartile range (IQR) of the speed modulus (i.e., rate) of all vehicles in the group, and excludes those whose rates are out of range (first quartile - 1.5 × IQR, third quartile + 1.5 × Vehicles with an IQR (interquartile range) are marked as speed outliers. Similarly, the same outlier detection is performed on the acceleration modulus. These marked abnormal vehicles are temporarily excluded in the subsequent average value calculation. Then, the system calculates the average velocity vector and average acceleration vector of the group. The average velocity vector is calculated by adding the x components of the velocity vectors of all non-abnormal vehicles and dividing it by the total number of non-abnormal vehicles to obtain the x component of the average velocity. Similarly, the y component is processed in the same way to obtain the y component of the average velocity. These two components together constitute the average velocity vector of the group. The calculation process of the average acceleration vector is exactly the same as that of the average velocity vector, except that the processed data object is replaced by the acceleration vector component. The calculated average velocity vector and average acceleration vector of each group are associated and stored together with their corresponding group ID. This contains a set of collective motion state descriptors of all vehicle groups, which ultimately constitutes the regional motion state benchmark.

[0022] The steps for obtaining individual driving deviation values ​​are as follows: Based on the regional motion state benchmark, the velocity component and acceleration component corresponding to a single vehicle at each time step are extracted to construct the vehicle's four-dimensional motion vector at each time step. The four-dimensional motion vectors corresponding to all time steps are integrated to form a continuous vehicle motion vector sequence, thus obtaining a single vehicle motion vector sequence. According to the single vehicle motion vector sequence, the Mahalanobis distance between the vehicle motion vector at each time step and the average motion vector of the dynamic traffic micro-cluster to which it belongs is calculated to form the individual driving deviation value.

[0023] Specifically, based on the regional motion state benchmark, the system calls the unique tracking ID of each independently tracked vehicle in the real-time motion parameter set of the vehicle, continuously extracts its motion data at each time step from the set, the length of a time step is set to 50 milliseconds, which is consistent with the data update frequency of 20 Hz of the front sensor, at each time step, the system accurately extracts the longitudinal and lateral components of the vehicle's speed vector in the road coordinate system, and the longitudinal and lateral components of the acceleration vector in the same coordinate system, for example, at time step t, the system obtains the longitudinal speed of vehicle ID "A001" as 15.8 m / s, the lateral speed as -0.3 m / s, the longitudinal acceleration as 1.1 m / s², and the lateral acceleration as -0.2 m / s², then the system combines the four values in a predefined order, i.e. [longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration], into a four-dimensional array, which is the four-dimensional motion vector of the vehicle at time step t, i.e. [15.8, -0.3, 1.1, -0.2], the system maintains a first-in-first-out data queue for each vehicle, the length of the queue is determined according to a preset time period, which is set to 5 seconds, this period is determined according to the research on driving behavior analysis, which can capture meaningful driving patterns and respond to sudden situations in time, therefore, the data queue stores the four-dimensional motion vectors of the past 5 seconds, i.e. 100 continuous time steps, every time a new four-dimensional motion vector of a time step is calculated and added to the front end of the queue, the oldest vector at the end of the queue is removed, in this way, the system dynamically maintains an ordered set containing all motion states of the vehicle in the past 5 seconds for each vehicle, obtaining the single vehicle motion vector sequence.

[0024] Based on the sequence of single vehicle motion vectors generated in the previous step, the system processes each time step in the sequence and calculates the degree to which each four-dimensional motion vector deviates from the behavior benchmark of the cluster to which it belongs. Specifically, for the four-dimensional motion vector of a specific vehicle at time step t, the system first determines the ID of the dynamic traffic micro-cluster to which the vehicle belongs at that moment by querying the dynamic affiliation record between the vehicle and the cluster. Then, the system uses this cluster ID to retrieve the overall motion statistical characteristics of the cluster from the regional motion state benchmark, including the four-dimensional average motion vector and the covariance matrix of the four-dimensional motion vector. The four-dimensional average motion vector is obtained by averaging the four-dimensional motion vectors of all vehicles in the cluster, while the 4x4 covariance matrix is ​​obtained by calculating the variance of the four-dimensional motion vectors of all vehicles in the cluster in each dimension and the covariance between different dimensions. The covariance matrix describes the overall distribution of vehicle motion states within the cluster and the correlation between variables. For example, it reflects not only the degree of velocity dispersion but also the correlation pattern between velocity changes and acceleration changes. The system then uses the Mahalanobis distance to quantify the difference between the single vehicle motion vector and the cluster average motion vector. The calculation formula is: ,in, is the four-dimensional motion vector of a single vehicle at time step t, is the four-dimensional average motion vector of the cluster to which it belongs, It is the inverse matrix of the cluster covariance matrix. This calculation process first obtains the difference vector between the vehicle vector and the cluster average vector. Then, by multiplying its transpose on the left and the inverse of the covariance matrix on the right, the distance difference in the multidimensional space is converted into a dimensionless scalar value that takes into account the data distribution. This scalar value is the individual driving deviation value of the vehicle at the current time step.

[0025] The steps to obtain the vehicle behavior instability measure are: Based on the individual driving deviation value, the vehicle behavior instability metric is calculated using the following formula: ; in, is the vehicle behavior instability measure, is the total number of time steps in the time period, is the time step The Mahalanobis distance represents the degree of deviation between the vehicle motion vector and the average motion vector of the dynamic traffic micro-cluster under the covariance structure. is the time step The velocity vector modulus of represents the instantaneous speed of the vehicle. is the average velocity vector modulus of the dynamic traffic micro-cluster, indicating the average rate of the cluster, A constant positive number.

[0026] Specifically, the formula: The benefit of the formula is that it does not use the deviation of the multi-dimensional motion state (Mahalanobis distance) in isolation. ), but rather introduces a nonlinear weighting mechanism of speed difference through a hyperbolic tangent function (tanh), this weighting term When the instantaneous speed of a vehicle is significantly higher than the average speed of its cluster, its driving deviation value It will be amplified, and vice versa. This captures the fact that abnormal behaviors at high speed (such as sudden acceleration and lane change) have higher risks than similar behaviors at low speed. The design of the formula uses the relative size of speed as a regulating factor to make the final measurement value more sensitive to potentially dangerous high-speed unstable behaviors. In addition, the use of the root mean square structure to aggregate the weighted deviation values ​​within a time period T can smooth out instantaneous and accidental disturbances and obtain an instability score that is more representative and robust in time, rather than overreacting to single-point events. The final vehicle behavior instability measure is It can more accurately reflect the driver's continuous driving style and potential risk level.

[0027] The total number of time steps in a time period defines the time window size for calculating the vehicle behavior instability metric. The setting of its value requires a trade-off between the real-time nature of the evaluation and the stability of the results. A shorter time period can respond more quickly to changes in driving behavior, but may generate noise due to accidental driving operations; a longer time period can provide a more stable evaluation, but may have a delayed response to sudden risks. In this method, by analyzing a large amount of real traffic data streams containing normal driving and accident precursor events, it is found that a 5-second time window can achieve the best balance in most scenarios. Therefore, under the condition that the system's data acquisition frequency is 20 Hz (i.e., each time step is 0.05 seconds), The value of is set to 100 (i.e., 5 seconds / 0.05 seconds / step), which means that the vehicle behavior instability metric is based on the comprehensive evaluation results of the vehicle's past 100 consecutive motion states.

[0028] is the time step The Mahalanobis distance, which is calculated by the above steps, is a quantitative measure of the distance between a single vehicle and the time step The Mahalanobis distance is a measure of the difference between a vehicle's four-dimensional motion vector (including velocity and acceleration components) and the average motion state of the dynamic traffic micro-cluster to which it belongs. It is a dimensionless value that represents the statistical distance of the vehicle's motion state from the center of its cluster, taking into account the variance and correlation of the motion variables within the cluster. For example, in the calculation in the previous step, at time step t=1, the difference between the four-dimensional motion vector of a vehicle and the center of its cluster, after considering the covariance structure of the cluster's overall motion distribution, is calculated to be 2.1.

[0029] is the time step The velocity vector modulus represents the instantaneous speed of the vehicle at that moment. This value is directly obtained from the real-time motion parameter set of the vehicle. The two components of the velocity vector (longitudinal velocity and lateral velocity ) is calculated, and the calculation formula is: For example, at time step t=1, the sensor fusion system measures the longitudinal velocity of a vehicle to be 19.9 m / s and the lateral velocity to be 2.0 m / s. Then its instantaneous velocity is The calculation result is Meters / second.

[0030] is the average velocity vector modulus of the dynamic traffic micro-cluster to which it belongs, representing the overall average driving speed of the cluster to which the vehicle belongs. This value comes from the regional motion state benchmark. First, the average velocity vector of the cluster needs to be obtained from the benchmark. , which is obtained by averaging the x and y components of the velocity vectors of all vehicles in the cluster. Then, the modulus of the average velocity vector is calculated using the formula: For example, a cluster contains three vehicles with velocity vectors of (15.8 m / s, 0.1 m / s), (16.0 m / s, -0.1 m / s), and (16.2 m / s, 0.0 m / s). The x-component of the average velocity vector is (15.8+16.0+16.2) / 3 = 16.0 m / s, and the y-component is (0.1-0.1+0.0) / 3 = 0.0 m / s. Therefore, the average velocity vector is (16.0, 0.0), and its modulus is for Meters / second.

[0031] is a constant positive number, which is used to prevent the denominator from being zero in the calculation. Under certain traffic conditions, such as when the road is severely congested and the vehicle group is completely stationary, the average velocity vector modulus of the dynamic traffic micro-cluster is Possibly zero.

[0032] Computational procedure: Take the example of a single vehicle whose behavior is to be measured for instability over a period of 3 time steps (i.e. ). The detailed computational procedure is shown as follows: According to the previous steps, the average speed of the cluster to which the vehicle belongs has been obtained m / s, and the constant .

[0033] Over the successive 3 time steps, the instantaneous speed of the vehicle and the Mahalanobis distance are monitored and computed as follows: Time step t = 1 : the vehicle accelerates to overtake, m / s, .

[0034] Time step t = 2 : the vehicle returns to the main road, m / s, .

[0035] Time step t = 3 : the vehicle resumes cruising, m / s, .

[0036] The computational procedure is as follows: Compute the weighted deviation term for each time step .

[0037] For t = 1 : Speed difference term: ; Weighting factor: ; Weighted deviation term: ; For t = 2 : Speed difference term: ; Weighting factor: ; Weighted deviation term: ; For t = 3 : Speed difference term: ; Weighting factor: ; Weighted deviation term: ; Compute the root mean square of the weighted deviation terms.

[0038] Mean square sum: ; ; Prescription: ; The results show that the vehicle's behavior instability measure in these three time steps is The calculated value is 2.05, which comprehensively reflects the degree to which the vehicle's driving behavior deviates from the traffic flow benchmark during the observation period, and focuses on the additional risk brought by its speeding behavior. The system compares this value with the preset risk level threshold, which is obtained through statistical analysis of massive historical traffic data. For example, the analysis results show that the normal driving vehicle 95% of the values ​​are below 1.5, so 1.5 is set as the dividing line between stable and moderately unstable driving, and driving behaviors that indicate a precursor to an accident or dangerous event are also considered to be unstable. The values ​​are generally higher than 3.0, so 3.0 is set as the dividing line between moderate instability and high-risk driving. The current calculated value of 2.05 is between 1.5 and 3.0, indicating that the vehicle exhibits moderate behavioral instability and requires continuous attention from the system.

[0039] The steps for obtaining the cumulative amount of road driving risk are as follows: Based on the vehicle behavior instability measurement of all vehicles in the target road section, the vehicle behavior instability measurement values ​​of all vehicles in the current time step are extracted, and the extracted values ​​are summed up to obtain the cumulative driving risk of the current road section at the current time step.

[0040] Specifically, based on the vehicle behavior instability metric for all vehicles within a target road section, the system first defines the spatial scope of the target road section. For example, for a critical intersection, the section is defined as a 500-meter area extending along the road centerline from its upstream entrance. The system initiates a risk accumulation calculation at each time step, i.e., every 50 milliseconds. At this time step, the system first traverses its internally maintained set of real-time vehicle motion parameters. By comparing each vehicle's real-time location data with the preset section geofence coordinates, it filters out all vehicles currently traveling within the 500-meter section and obtains a list of their unique identification IDs. Based on this ID list, the system then queries and extracts the latest calculated vehicle behavior instability metric value for each vehicle from a dynamically updated data table. This table, with vehicle ID and timestamp as the joint primary key, stores all historical and current metric values ​​for all vehicles. The extraction operation returns a list of values. For example, at the current time step, there are 35 vehicles in the section, and the system extracts these 35 metric values, such as [2.05, 0.8, 1.7, ..., 1.1], then the system performs a summation process on this list of 35 values, that is, adding up all the values. For example, if the sum of these 35 vehicle behavior instability metric values ​​is 48.6, then this value is confirmed as the cumulative driving risk of the current road section at the current time step.

[0041] The steps to obtain the traffic flow conflict risk index are as follows: According to the cumulative driving risk of the road section, the real-time traffic volume, road capacity and average speed of the road section at the corresponding time step are retrieved and included in the calculation input set together with the cumulative driving risk of the road section to obtain the joint variable set that constitutes the risk index calculation; Based on the joint variable set, the traffic flow conflict risk index is calculated using the following formula: ; in, is the traffic flow conflict risk index, is the cumulative driving risk of the road section, is the real-time traffic flow of the specified road section at the current moment, is the traffic capacity of a specified road section, is the average vehicle speed of a specified road section, The optimal speed set for the section, is the congestion sensitivity factor, is the speed deviation sensitivity factor.

[0042] Specifically, based on the cumulative driving risk of the current road section at the current time step obtained in the previous step, the system synchronously retrieves the three macro traffic flow parameters corresponding to the section and the time step, namely, real-time traffic volume, road capacity and average speed. Among them, the real-time traffic volume is counted by setting a virtual detection line at the end of the section. The system counts the total number of vehicles passing through the detection line in the past 5 minutes, and then extrapolates this count value to the hourly traffic volume. For example, if 150 vehicles passed in the past 5 minutes, the real-time traffic volume is calculated as 150 vehicles / 5 minutes × 60 minutes / hour = 1,800 vehicles / hour, and road capacity is a static parameter preset in the system's road network database. It is calculated based on the physical properties of the section, such as the number of lanes, lane width, lateral clear width, and road slope, according to the standard method in the Highway Capacity Manual (HCM). For example, the theoretical capacity of a two-lane urban trunk road with a design speed of 80 kilometers per hour is set at 3,600 vehicles per hour. Finally, the average vehicle speed is obtained by calculating the average instantaneous speed of all vehicles in the section at the current time step. For example, if the average instantaneous speed of 35 vehicles in the section is 62 kilometers per hour, then the average vehicle speed is 62 kilometers per hour. The system packages these three macro parameters together with the previously obtained cumulative driving risk of the road section (for example, 48.6) to form a data structure containing four key variables, resulting in a joint variable set that constitutes the risk index calculation.

[0043] formula: , the formula is beneficial in that the exponential amplification term Two key macro traffic flow adjustment factors are introduced: traffic congestion level (determined by real-time traffic volume) and road capacity ) and traffic flow speed stability (expressed by the average vehicle speed and optimal speed The use of the exponential function makes the risk index The growth of congestion and speed deviation shows a nonlinear and exponential amplification effect, which is consistent with the phenomenon that the risk of accidents increases sharply when traffic flow approaches the collapse point in reality. and The model can be calibrated according to the characteristics of different road types, and the final output traffic flow conflict risk index is It not only quantifies the current risk level, but also determines whether the risk is dominated by individual driving behavior or by the deterioration of the overall traffic flow status.

[0044] is the cumulative driving risk of the road section. This parameter is directly derived from the calculation results of the previous steps. It is obtained by summing the vehicle behavior instability measures of all vehicles in the target road section at a specific time step. It quantifies the basic risk level constituted by the unstable behaviors of individual drivers on the road section at that moment. The higher the value, the more or more severe unstable driving behaviors there are in the road section. For example, in the previous step, the system identified 35 vehicles in the target section and extracted their respective vehicle behavior instability measures. By summing up these 35 values, the cumulative driving risk of the road section at the current time step is obtained. It is 48.6.

[0045] The real-time traffic volume at a given road section at the current moment, measured in vehicles per hour (vph), is calculated by collecting traffic volume within a fixed time window using sensors (or virtual coils) deployed at the end of the road section. The collection time window is typically set to 5 minutes to balance the real-time nature and smoothness of the data. For example, if the system detects that 150 vehicles have passed through the exit monitoring point of the section in the past 5 minutes, the real-time traffic volume at the current moment is calculated as: vehicles / hour.

[0046] The capacity of a specified road section is expressed in vehicles per hour (vph). This is a static parameter pre-calibrated according to road design standards and physical conditions and stored in the system's road network configuration database. The calibration process follows the standard specifications of traffic engineering and takes into account factors such as the number of lanes, lane width, shoulder width, road longitudinal slope, proportion of large vehicles, and lateral interference. For example, for a two-lane, one-way urban trunk road section with a design speed of 80 km / h, no obvious slope, and no interference such as bus stops, the basic capacity of a single lane is 1900 vehicles per hour. Therefore, the total capacity of the two-lane section is calibrated as vehicles / hour.

[0047] The average speed of a specified road section, in kilometers per hour (km / h), is obtained by calculating the average instantaneous speed of all vehicles in the section at the current time step. It reflects the real-time macroscopic speed of traffic flow. The system first obtains the instantaneous speed set of all vehicles in the section and then calculates the average value. For example, at the current time step, the system monitors a total of 35 vehicles in the section. By averaging the instantaneous speeds of these 35 vehicles, the average speed of the section is obtained. km / h.

[0048] The optimal speed set for the section is in kilometers per hour (km / h). This speed is the speed at which the traffic flow of the road section reaches its maximum (i.e., reaches its capacity). It is not the speed limit of the road. It is obtained by statistically analyzing the long-term historical traffic data of the section (speed-flow relationship). The specific method is to draw a speed-flow scatter plot and fit a curve that can describe the relationship between the two (such as the Greenberg model or the Underwood model). The speed corresponding to the peak point of the flow on the curve is the optimal speed. For example, by analyzing the historical data of the target section for three consecutive months, it is found that when the average speed is around 65 kilometers per hour, the hourly flow of the section is the highest. Therefore, the optimal speed of the section is set to km / h.

[0049] The congestion sensitivity factor is a dimensionless weight coefficient used to adjust the sensitivity of the traffic flow conflict risk index to the degree of traffic congestion (i.e., the ratio of traffic volume to traffic capacity). The value of this factor is calibrated by regression analysis of historical data. The analysis data includes historical records of traffic congestion levels and the frequency of traffic accidents or serious traffic conflicts that occurred during the same period. For those sections of road where congestion is very likely to cause chain reactions and accidents once it forms (such as ramp merging areas), The value of will be higher. The calibration process is to build a logistic regression model with congestion as the independent variable and accident rate as the dependent variable. The coefficient of the congestion term in the model is The reference value of the congestion sensitivity factor is set. After analyzing the historical accident data of the target road section, it is found that for every 0.1 increase in congestion, the logarithmic occurrence ratio of accident risk increases by 0.15. .

[0050] is the speed deviation sensitivity factor, which is a dimensionless weight coefficient used to adjust the sensitivity of the traffic flow conflict risk index to the degree to which the average speed deviates from the optimal speed. This ensures that both slower than optimal speeds (which often means increased congestion) and faster than optimal speeds (which mean reduced safety margins) increase risk. The calibration process and Similarly, by analyzing the relationship between historical speed data and accident data, for sections of highways that require high speed uniformity, The value of will be higher. After analyzing the historical data of the target road section, it is found that there is a stronger correlation between speed deviation and accident risk. Therefore, the speed deviation sensitivity factor is set to .

[0051] Calculation process: An example calculation is performed with the aforementioned parameter values plugged into the formula, where , , , , , , .

[0052] The congestion term in the index portion is calculated as: ; The speed deviation term in the index portion is calculated as: ; The complete index value is calculated as: ; The index amplification factor is calculated as: ; The final traffic flow conflict risk index is calculated as : ; The result shows that the current road section traffic flow conflict risk index is 99.33, which is a comprehensive risk score that combines the microscopic individual driving instability (with a base risk value of 48.6) with the macroscopic traffic flow state (with congestion and speed deviation jointly amplifying the base risk by about 2.04 times). The system compares this result with the pre-set multi-level risk threshold, for example, the preset threshold is: low risk (0-50), medium risk (50-100), high risk (100-200), and extremely high risk (>200). The current calculated 99.33 is at the upper limit of the medium risk interval and close to the high risk level.

[0053] The preset control strategy set acquisition step is: Based on the road section traffic flow conflict risk index, a pre-defined multi-level risk threshold set is called, each risk threshold in the risk threshold set is extracted one by one, the traffic flow conflict risk index is compared with each risk threshold in terms of numerical value size, the risk level matching the current traffic flow conflict risk index is determined, and a road section risk level determination result is generated; According to the road section risk level determination result, the risk level determination result is taken as a query condition, the pre-prepared control strategy set corresponding to the risk level is called from the associated control strategy library, and a preset control strategy set suitable for the current road section risk level is formed.

[0054] Specifically, based on the traffic flow conflict risk index of the road section, the system calls a predefined multi-level risk threshold set, which is constructed based on the comprehensive analysis of historical traffic data, accident data and traffic simulation results of the specific road section. Specifically, first, collect all traffic flow conflict risk index calculation values of the section at all time points in the past year to form a large historical index data set. At the same time, collect traffic accident reports during the same period, and mark the risk index values 5 minutes before each accident as high-risk samples. Then, use the percentile method to divide the historical index data set, for example, the 70th percentile of the data set is taken as the boundary between "low risk" and "medium risk", the 90th percentile is taken as the boundary between "medium risk" and "high risk", and the 98th percentile is taken as the boundary between "high risk" and "extremely high risk". Adjust these thresholds in combination with the distribution of accident samples to ensure that most accident samples fall into the high-risk or extremely high-risk range. Finally, a clear threshold set is formed, for example: ["low risk": (0, 50], "medium risk": (50, 100], "high risk": (100, 200], "extremely high risk": (>200)]. The system compares the currently calculated traffic flow conflict risk index, for example 99.33, with the thresholds in this set one by one. The comparison process starts from the lowest risk level. First, determine whether 99.33 is greater than 0 and less than or equal to 50. The result is no. Then determine whether it is greater than 50 and less than or equal to 100. The result is yes. Once the matching interval is found, the comparison process terminates, and the system determines that the current risk level is "medium risk", and generates the road section risk level determination result.

[0055] Based on the road section risk level determination result generated in the previous step, for example, "medium risk", the system uses this determination result as the only query keyword to access a built-in associated control strategy library. This library is a structured database that pre-stores control strategy sets for different risk levels, different road types, and different time periods (such as peak, off-peak, and nighttime). These strategy sets are jointly developed by a team of traffic engineers and traffic management experts based on traffic engineering theory, simulation optimization results, and management practice experience. Each strategy has been rigorously demonstrated and tested. For example, the library stores a strategy set corresponding to the "medium risk" level, which may include the following: Content: Strategy 1: For downstream traffic lights, the instruction is "extend the cycle by 10%", and the parameter is "the green signal ratio is tilted 5% toward the main road"; Strategy 2: For upstream variable information signs, the instruction is "display text", and the parameter is "traffic is dense ahead, please slow down"; Strategy 3: For the traffic broadcast system interface, the instruction is "report road conditions", and the parameter is "There is a moderate congestion risk on the XX road section, and it is recommended to detour". When the system performs a query operation, it uses "medium risk" as an index to accurately match and retrieve this set of multiple specific response measures from the library. This set is a preset control strategy set customized for the current situation and composed of multiple sub-strategies.

[0056] The steps for obtaining adaptive traffic control instructions are: Based on the preset control strategy set, strategy instructions are extracted one by one from the control strategy set and converted into instruction data streams that can be executed by traffic control equipment. According to the instruction data streams, control actions are generated one by one for the traffic lights at the downstream intersection or the variable information signs ahead to form adaptive traffic control instructions.

[0057] Specifically, based on the preset control strategy set that has been obtained and is applicable to the current road section risk level, the system begins to parse and execute the strategy instructions one by one, converting the high-level strategy description into specific instructions that the underlying traffic control equipment can recognize and execute. For the first strategy in the preset control strategy set, "For downstream traffic lights, the instruction is 'extend the cycle by 10%', and the parameter is 'tilt the green light ratio towards the main road by 5%'", the system first connects to the traffic signal controller at the downstream intersection and queries its current signal timing plan. For example, the current plan is a 120-second cycle and a 50-second green light time for the main road. The system calculates the new timing parameters based on the instruction: the new cycle is 120 seconds × (1 + 10%) = 132 seconds, and the new green light time for the main road is 50 seconds × (1 + 5%) = 52.5 seconds (rounded to 53 seconds), and then encapsulates these new parameters into a command data packet that complies with the communication protocol of this model of signal (such as the NTCIP protocol) and sends it to the signal through the network. For the second strategy "For the upstream variable message sign, the instruction is 'display text' and the parameter is 'Traffic ahead is dense, please slow down'", the system connects to the variable message sign (VMS) controller at the specified location, encodes the text parameter "Traffic ahead is dense, please slow down" into a format supported by the VMS controller (for example, a specific encoding containing information such as font, color, and display mode), generates a control data stream and sends it. After receiving it, the VMS will update the display content. The system processes all policy instructions in sequence according to the order in which the policies are concentrated, generates and sends corresponding control actions for each target device, and these actions together constitute real-time, closed-loop adaptive traffic control instructions for the current traffic conditions.

[0058] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A traffic monitoring method for vehicle tracking and early warning, characterized in that: The following steps are involved: Obtain real-time position, speed, and acceleration data for all vehicles within the monitored road area. Based on the spatial distance between vehicles and the time window setting, adjacent vehicles are aggregated into dynamic traffic micro-clusters to establish a regional motion state benchmark. Based on the regional motion state benchmark, a motion vector of a single vehicle is extracted to calculate an individual driving deviation value, and then the individual driving deviation values ​​of the single vehicle are summed up over a set time period to obtain a vehicle behavior instability measure; Based on the vehicle behavior instability measure of each vehicle, the vehicle behavior instability measures of all vehicles on the target road section are summed to obtain a cumulative driving risk of the road section, and the cumulative driving risk of the road section is then calculated with the real-time traffic volume, average speed and road capacity of the road section to generate a traffic flow conflict risk index for the road section; Based on the traffic flow conflict risk index of the road section, the traffic flow conflict risk index is compared one by one with the preset multi-level risk threshold value, and according to the comparison result, a preset control strategy set is matched and selected from the associated control strategy library. According to the preset control strategy set, adaptive traffic control instructions are generated for the traffic lights at the downstream intersection or the variable information signs ahead.

2. The traffic monitoring method for vehicle tracking and early warning according to claim 1, characterized in that: The steps for obtaining the regional motion state benchmark are: Through the video sensors and millimeter-wave radar deployed in the road area, all vehicles currently traveling in the road area are detected synchronously and in real time, and the real-time position data, speed data and acceleration data of each vehicle are obtained. The data of each vehicle is uniformly marked with a timestamp to generate a set of real-time vehicle motion parameters. Based on the set of real-time vehicle motion parameters, according to a preset inter-vehicle spatial distance threshold and a time window threshold for vehicle motion state collection, vehicles whose spatial distance is below the spatial distance threshold and whose data collection time difference is within the time window threshold are identified as adjacent vehicles, and the adjacent vehicles are aggregated into multiple vehicle groups to generate dynamic traffic micro-clusters; Based on the dynamic traffic micro-cluster, speed data and acceleration data of all vehicles in each dynamic traffic micro-cluster are extracted to obtain a regional motion state benchmark.

3. The traffic monitoring method for vehicle tracking and early warning according to claim 1, characterized in that: The steps for obtaining the individual driving deviation value are as follows: Based on the regional motion state benchmark, the velocity component and acceleration component corresponding to a single vehicle at each time step are extracted, a four-dimensional motion vector of the vehicle at each time step is constructed, and the four-dimensional motion vectors corresponding to all time steps are integrated to form a continuous vehicle motion vector sequence to obtain a single vehicle motion vector sequence; According to the single vehicle motion vector sequence, the Mahalanobis distance between the vehicle motion vector at each time step and the average motion vector of the dynamic traffic micro-cluster to which it belongs is calculated to form an individual driving deviation value.

4. The traffic monitoring method for vehicle tracking and early warning according to claim 1, characterized in that: The steps for obtaining the vehicle behavior instability measure are: Based on the individual driving deviation values, a vehicle behavior instability metric is calculated.

5. The traffic monitoring method for vehicle tracking and early warning according to claim 1, characterized in that: The steps for obtaining the cumulative amount of driving risk on the road section are as follows: Based on the vehicle behavior instability measurement of all vehicles in the target road section, the vehicle behavior instability measurement values ​​of all vehicles in the current time step are extracted, and the extracted values ​​are summed up to obtain the cumulative driving risk of the current road section in the current time step.

6. The traffic monitoring method for vehicle tracking and early warning according to claim 1, characterized in that: The steps for obtaining the traffic flow conflict risk index are: According to the cumulative driving risk of the road section, the real-time traffic volume, road capacity and average speed of the road section at the corresponding time step are retrieved and included in the calculation input set together with the cumulative driving risk of the road section to obtain a joint variable set constituting the risk index calculation; Based on the joint variable set, a traffic flow conflict risk index is calculated.

7. The traffic monitoring method for vehicle tracking and early warning according to claim 1, characterized in that: The steps for obtaining the preset control strategy set are: Based on the traffic flow conflict risk index of the road section, a predefined multi-level risk threshold set is called, each risk threshold in the risk threshold set is extracted one by one, the traffic flow conflict risk index is compared with each risk threshold item by item, the risk level that matches the current traffic flow conflict risk index is determined, and a road section risk level determination result is generated; According to the road section risk level determination result, the risk level determination result is used as a query condition, and a pre-established control strategy set corresponding to the risk level is retrieved from the associated control strategy library to form a preset control strategy set suitable for the current road section risk level.

8. The traffic monitoring method for vehicle tracking and early warning according to claim 1, characterized in that: The steps for obtaining the adaptive traffic control instruction are: Based on the preset control strategy set, strategy instructions are extracted one by one from the control strategy set and converted into instruction data streams that can be executed by traffic control equipment. According to the instruction data streams, control actions are generated one by one for the traffic lights at the downstream intersection or the variable information signs ahead to form adaptive traffic control instructions.

9. The traffic monitoring system for the traffic monitoring method for vehicle tracking and early warning according to any one of claims 1 to 8, characterized in that: include: The data acquisition module obtains the real-time position, speed, and acceleration data of all vehicles in the monitored road area. Based on the spatial distance between vehicles and the time window setting, adjacent vehicles are aggregated into dynamic traffic micro-clusters to establish a regional motion state benchmark; a behavior assessment module that extracts a motion vector of a single vehicle based on the regional motion state benchmark, calculates an individual driving deviation value, and then sums the individual driving deviation values ​​of the single vehicle over a set time period to obtain a vehicle behavior instability measure; a risk calculation module that sums the vehicle behavior instability metrics of all vehicles on a target road section based on the vehicle behavior instability metrics of each vehicle to obtain a cumulative driving risk for the road section, and then calculates the cumulative driving risk for the road section with the real-time traffic volume, average speed, and road capacity of the road section to generate a traffic flow conflict risk index for the road section; The control instruction module compares the traffic flow conflict risk index of the road section with the preset multi-level risk threshold one by one, matches and selects the preset control strategy set from the associated control strategy library according to the comparison result, and generates adaptive traffic control instructions for the traffic lights at the downstream intersection or the variable information signs ahead according to the preset control strategy set.

Images