Shared bicycle dangerous behavior identification method based on GPS trajectory data tracking

Abstract

The invention discloses a shared bicycle dangerous behavior identification method based on GPS trajectory data tracking, and belongs to the technical field of traffic safety. Comprising the steps of abnormal point elimination, point supplementation, coordinate system unification and normalization so as to improve data accuracy, and clustering division is carried out on data according to data features; in combination with a road network and POI data, hotspot and non-hotspot areas are determined through track intensity analysis; six dangerous behaviors are defined, historical data are analyzed, a threshold value is set, a predefined method is used for recognizing dangerous driving behaviors, and finally a recognition model is established to achieve accurate recognition of the six dangerous behaviors. Through mathematical statistics, the riding characteristics and dangerous behavior characteristics of the shared bicycle under different conditions are compared and researched; according to the dangerous behavior identification method based on the shared bicycle GPS track data, the utilization rate of the shared bicycle GPS data can be effectively improved, dangerous behaviors can be accurately identified, and the safety of riders and other traffic participants is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical fields of traffic safety and data analysis, and particularly relates to a method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking. Background Art

[0002] As an important promoter for promoting green travel, reducing traffic congestion and environmental pollution, shared bicycles are also an important guarantee for solving the last-mile problem of urban transportation. Now they have been well-known to the general public and become a means of transportation for daily use. However, with the increasing number of users of shared bicycles, more and more risks of riding shared bicycles begin to affect people's lives and property safety, and uncivilized behaviors also pose hidden dangers to traffic safety. Different from floating cars, etc., shared bicycles can use on-vehicle GPS to detect the dynamics of vehicles in real time, and shared bicycles are small and can enter many roads that are not counted in the road network data, facing more complex road conditions. Therefore, further standardizing the behaviors of shared bicycles is very important for urban traffic safety and the lives and property safety of residents. Summary of the Invention

[0003] To solve the above problems, the present invention discloses a method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking, which effectively improves the utilization rate of GPS data of shared bicycles, accurately identifies dangerous behaviors, and ensures the safety of riders and other traffic participants.

[0004] To achieve the above object, the technical solution of the present invention is as follows:

[0005] A method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking includes the following steps:

[0006] S1 Preprocess the GPS trajectory data of shared bicycle orders.

[0007] S2 Define six dangerous behaviors of shared bicycles: dangerous speeding behavior, dangerous sudden deceleration behavior, dangerous parallel riding behavior, running a red light behavior, crossing the road behavior, and following a vehicle to turn left behavior.

[0008] S3 Determine the thresholds of six dangerous driving behaviors by analyzing historical data, including but not limited to speeding threshold, red light running threshold, sudden deceleration threshold, collision time threshold, dangerous parallel riding threshold, direction angle threshold, and left turn distance threshold.

[0009] S4 Traverse the data of all shared bicycle orders and use a predefined method to identify dangerous driving behaviors.

[0010] Further, the method includes the following steps: traversing the shared bicycle order trajectory data set, removing trajectory points with errors exceeding a specific threshold due to weak GPS signals; judging jump points according to time difference and position, calculating the shortest weighted path between jump points using the Floyd algorithm, and filling in points; interpolating the trajectory points using the spline interpolation method to densify all order trajectory points; converting the coordinate system of the trajectory point coordinates into the UTM coordinate system and normalizing the time involved in the data; dividing the shared bicycle data within the same time window with a 30-second time window; calculating the average speed of each order data, and allocating time to each trajectory point according to the average speed to ensure that each trajectory point has a corresponding time; traversing the shared bicycle trajectory data with the order start time as the index, using the DBSCAN clustering algorithm, and dividing the order trajectory data into multiple clustering sets according to the spatio-temporal characteristics of the trajectory and the similarity of the time window; considering the distance between trajectory points, the time stamp difference, and the number of core points in each cluster during the clustering process; importing road network data and POI data obtained from public data sources, performing road network matching, and extracting hot spots and non-hot spots according to the trajectory density.

[0011] Further, the method also includes: when the relative speed of the shared bicycle is greater than the overspeed threshold, and the lateral distance between the trajectory point of the bicycle and the adjacent trajectory point is less than 0.5 m or the adjacent trajectory has an avoidance behavior, it is determined as a dangerous overspeed behavior; when the deceleration of the shared bicycle is less than the emergency deceleration threshold, it is identified as a deceleration trajectory, and the deceleration starting point is defined. At the same time stamp, calculate the collision time between the deceleration starting point and the adjacent trajectory point of the following vehicle. When the collision time is less than 1 second, it is determined that the leading vehicle has a dangerous emergency deceleration behavior; set a red light waiting area. When the shared bicycle crosses the red light waiting area and continues to ride, compare its direction angle. If the direction angle is less than the threshold, it is determined as a red light running behavior; when the Euclidean distance between shared bicycles with similar trajectories and the same time window is less than the dangerous parallel riding threshold, it is identified as a parallel behavior. When the difference between the lateral distance of the dangerous parallel riding vehicle and the road width is less than 1 m, and the subsequent other behaviors of the parallel vehicles at the same time stamp are similar to the parallel trajectory at the same time stamp, it is determined that the leading vehicle has a dangerous parallel behavior; when the trajectory point deviates from the original road and there is a large change in the direction angle of the segmented trajectory where the trajectory point is located, the change in the direction angle exceeds the direction angle threshold, and the displacement of the starting point and the ending point of the segmented trajectory on the coordinate axis perpendicular to the road is greater than half of the road width, it is determined as a behavior of crossing the road; segment the trajectory within the crossroads area separately. The direction angles of the starting point and the ending point of the segmented trajectory are greater than It is identified as a left-turn trajectory. If there are no stopping points among the trajectory points of the left-turn trajectory except the starting point and the ending point, the direction angle of the trajectory points is increasing steadily, and the length of the trajectory is less than the left-turn distance threshold, it is determined as a behavior of following a vehicle to turn left.

[0012] Further, the method further includes that the speed of the shared bicycle is V n，t :

[0013]

[0014] where n is the order number; t is the sampling time of the first sampled trajectory point; T is the sampling time interval between adjacent trajectory points; (long n,t+T , lat n,t+T ) is the longitude and latitude of the coordinate of the trajectory point at time t + T of the nth trajectory; S n,t is the straight-line distance between adjacent trajectory points of the same trajectory, representing the distance traveled by the shared bicycle from time t to time t + T.

[0015] Further, the method further includes that the speed of the shared bicycle is the deceleration

[0016]

[0017] where V n，t+T is the speed of the shared bicycle at the trajectory point at time t + T of the nth trajectory; S n,t is the distance between adjacent trajectory points of the same trajectory.

[0018] Further, the method further includes that the direction angle α of the shared bicycle n,t (in radians):

[0019] α n,t = |atan2(x, y)|

[0020] y = sin(Δlong)·cos(lat n,t+T )

[0021] x = cos(lat n,t )sin(lat n,t+T ) - sin(lat n,t )cos(lat n,t+T )cos(Δlong)

[0022] Δlong = long n,t+T - long n,t

[0023] where, (long n,t+T , lat n,t+T ) is the longitude and latitude of the coordinate of the trajectory point at time t + T of the nth trajectory. Δlong is the difference in the abscissa of the trajectory points.

[0024] Further, the method further includes that the polar radius ρ of the shared bicycle n,t :

[0025]

[0026] where a = 0.5 m, which is the semi - minor axis of the ellipse; b = 1.5 m, which is the semi - major axis of the ellipse; α n,t is the direction angle, converted to radians.

[0027] Furthermore, the method further includes the time - to - collision TTC of shared bicycles nm,t :

[0028]

[0029] △V nm,t = V n,t - V n+m,t

[0030]

[0031] where X nm,t is the distance between the deceleration starting point of deceleration trajectory n and the trajectory point of subsequent trajectory m at the same timestamp, and △V nm,t is the relative velocity between the deceleration starting point of deceleration trajectory n and the trajectory point of subsequent trajectory m at the same timestamp; (long n+m,t , lat n+m,t ) are the coordinates of the trajectory point of decelerating vehicle n and the trajectory point of subsequent vehicle m at time t.

[0032] Furthermore, the method further includes that the method for identifying the behavior of running a red light includes the following steps:

[0033] S4.1.1 Traverse the shared - bicycle trajectory data, and use the DBSCAN clustering algorithm to divide the order trajectory data into multiple clustering sets according to the spatio - temporal characteristics of the trajectory and the similarity of the time window. During the clustering process, consider the distance between trajectory points, the difference in timestamps, and the number of core points in each cluster.

[0034] S4.1.2 Import the road network data with cross - walk coordinates and match the road network data with the shared - bicycle GPS trajectory data;

[0035] S4.1.3 Traverse the order trajectory data in each clustering set, and identify the red - light waiting areas by analyzing the density of trajectory points within a specific time interval. These areas are defined as geographical areas where trajectory points gather and stay for more than a preset threshold during the red - light period.

[0036] S4.1.4 For the trajectories that extend beyond the red-light waiting area, analyze the changes in their coordinate positions and direction angles. If, without reasonable stops, the change in the direction angle of the trajectory points does not exceed the set turning threshold and does not match the traffic signal phase, it is determined as a red-light running behavior. If it exceeds the set turning threshold, it is considered a turning behavior. Set the lower and upper limits of the turning threshold according to the actual situation as

[0037] S4.1.5 The identification of red-light running behavior ends.

[0038] Furthermore, the method further includes that the method for identifying dangerous speeding behaviors includes the following steps:

[0039] S4.2.1 Traverse all the shared bicycle trajectory data after data preprocessing, obtain the position changes and corresponding time intervals of all trajectory points, calculate the vehicle speeds of all trajectory points and use them as the instantaneous speeds of these trajectory points, and sort the speeds from smallest to largest.

[0040] S4.2.2 Introduce POI data. Specifically, obtain the POI information of a specific area through a public data source, and extract the 80th percentile of the trajectory point speeds in the hot spot area during the specified time period as the speeding threshold for this area during this time period; in non-hot spot areas, also extract the 80th percentile of the trajectory speeds as the speeding threshold, so that the threshold decreases in places with large traffic flow and increases in places with small traffic flow, which is in line with the actual situation;

[0041] S4.2.3 When the vehicle speeds of consecutive trajectory points in the trajectory dataset exceed the relative speed threshold and the duration of this behavior is greater than 3 seconds, identify these shared bicycle driving behaviors as suspected dangerous speeding behaviors;

[0042] S4.2.4 Traverse the suspected dangerous speeding behaviors. Define that when the distance between the shared bicycle trajectory points and the road center line increases, that is, when deviating from the road center line, it is a riding behavior of pulling over to avoid.

[0043] S4.2.5 Traverse the set of suspected dangerous speeding behaviors, calculate the Euclidean distance between the trajectory points of the suspected dangerous speeding behaviors and the normal behaviors at the same time stamp. When the trajectory points of the suspected dangerous speeding behaviors exceed the trajectory points of the normal behaviors, if the distance between the trajectory points of the suspected dangerous speeding behaviors and the normal behavior points is less than 0.5 meters, or the trajectory points of the normal behaviors have an avoidance behavior, then determine that this suspected dangerous speeding behavior is a dangerous speeding behavior.

[0044] S4.2.6 The judgment of dangerous speeding behavior ends.

[0045] Furthermore, the method further includes that the method for identifying dangerous sudden deceleration behaviors includes the following steps:

[0046] S4.3.1 Traverse the shared bike order trajectory dataset after data preprocessing, introduce POI data, specifically obtain POI information of a specific area through a public data source, calculate the deceleration of the trajectory interval, and sort from large to small;

[0047] S4.3.2 And extract the 90th percentile of the deceleration of the trajectory points in the hot and non-hot regions within each time period as the sharp deceleration threshold for that region and time period;

[0048] S4.3.3 Traverse each trajectory in all clustering sets. When the deceleration of consecutive trajectory points in the trajectory is less than the sharp deceleration judgment threshold, it is identified as a suspected dangerous sharp deceleration behavior;

[0049] S4.3.4 Define the starting trajectory point where the deceleration in the trajectory interval exceeds the sharp deceleration judgment threshold as the deceleration starting point. Traverse the set of suspected sharp deceleration trajectories, extract the trajectory points of the decelerating trajectories and the trajectory points of other normal driving trajectories in the clustering set, and calculate the vehicle distance Δd nm,t and the relative speed ΔV nm,t , and further calculate the time to collision, that is, Time To Collision. When TTC nm,t is less than 1 second, it is determined that the vehicle in front has a dangerous sharp deceleration behavior;

[0050] S4.3.5 Determine the end of the dangerous sharp deceleration behavior.

[0051] Furthermore, the method further includes that the method for identifying dangerous parallel behaviors includes the following steps:

[0052] S4.4.1 Perform DBSCAN clustering analysis on the data to obtain clustering sets with the same time window, high trajectory similarity, and similar speeds, and verify the spatial position of the trajectory in combination with road network data;

[0053] S4.4.2 Traverse each clustering set, calculate the Euclidean distance d nm,t between the corresponding trajectory points of adjacent order trajectory data, and record the position of the trajectory points in the road network;

[0054] S4.4.3 Traverse the corresponding trajectory points of adjacent order trajectory data in each clustering set. Taking a certain point as the coordinate origin, calculate the direction angle α n,t of the corresponding trajectory points, and analyze the rationality of this direction angle in combination with road network data;

[0055] S4.4.4 Taking the current trajectory point as the origin, with a semi-minor axis of 0.5 meters and a semi-major axis of 1.5 meters, calculate the polar radius ρ n,t of the direction angle of the corresponding trajectory point, as the parallel threshold for this direction angle, and record the parallel threshold under this direction angle to simplify the calculation;

[0056] S4.4.5 Traverse each clustering set, analyze the changes in coordinate positions and timestamps. If the Euclidean distance between the trajectory points corresponding to adjacent order trajectories is less than the parallel threshold, it is identified as a parallel behavior, and the rationality of this behavior is verified by combining road network data;

[0057] S4.4.6 Calculate the lateral distance X of the vehicle with a dangerous parallel behavior in the direction perpendicular to the road n,t ; If the difference between the lateral distance X n,t and the width d of the road is less than 1 meter, and the subsequent other behaviors of the parallel vehicles are similar to the parallel trajectory within the same timestamp, it is determined that the leading vehicle has a dangerous parallel behavior;

[0058] S4.4.7 Determine the end of the dangerous parallel behavior.

[0059] Furthermore, the method further includes a method for identifying the behavior of crossing the road, which includes the following steps:

[0060] S4.5.1 Import the non-motor vehicle road network data of the data collection location, traverse the GPS trajectory data of the shared bicycle orders, and perform road network matching; extract the trajectories with 3 or more trajectory points deviating from the original road and the coordinate change in the direction perpendicular to the road exceeding half of the road width to obtain the set of trajectories to be detected;

[0061] S4.5.2 Traverse the set of trajectories to be detected, calculate the direction angle of the trajectory points, and segment the offset trajectory points to obtain trajectory offset segments. Compare the direction angles of the trajectory points before and after the trajectory offset segments. According to historical data analysis, when the change range of the direction angle is within ±8°, the trajectory points are considered straight trajectory points. Extract five straight trajectory points before and after the trajectory offset segments respectively. When the number of trajectory points is insufficient, extract the actual number of straight trajectory points to obtain the set of segmented trajectories to be detected;

[0062] S4.5.3 Traverse the set of segmented trajectories to be detected, calculate the direction angle of the trajectory points and the displacement of the horizontal and vertical coordinates of the starting and ending points of the segmented trajectories. If there is a change in the direction angle of the trajectory points greater than and the displacement of the horizontal and vertical coordinates of the starting and ending points of the segmented trajectories is greater than half of the average road width within the sampling time interval T, and the straight-line distance between the first offset point of the trajectory and the coordinates of the zebra crossing > 10m, then it is determined that the trajectory is a behavior of crossing the road;

[0063] S4.5.4 Identify the end of the behavior of crossing the road.

[0064] Furthermore, the method further includes a method for identifying the behavior of following the vehicle to turn left, which includes the following steps:

[0065] S4.6.1 Use the coordinates of the crosswalk at the intersection as the boundary, traverse the GPS trajectory data of all orders, extract the segmented trajectories of the orders passing through the intersection within the boundary of the intersection, and obtain a set of segmented trajectories;

[0066] S4.6.2 Traverse all the order trajectories in the set of trajectories to be judged. For each trajectory, calculate the direction angles of multiple key trajectory points on the trajectory, including the trajectory point just entering the intersection, several intermediate points, and the trajectory point after exiting the intersection; by analyzing the change trend of these direction angles, if the change amount of the direction angle after exiting the intersection compared with when entering is greater than then it is identified as a left-turn trajectory, thus obtaining a set of left-turn trajectories;

[0067] S4.6.3 Traverse all the segmented trajectories, calculate the speeds of the trajectory points in the segmented trajectories. If there are stopping points in the trajectory, and the direction angle of the trajectory points remains stable before the stopping point and steadily increases after passing the stopping point, then take it as a normal left-turn trajectory; arrange the distances of the normal left-turn trajectories from small to large;

[0068] S4.6.4 Take the minimum value as the left-turn distance threshold S min ;

[0069] S4.6.5 Traverse the distances of the remaining all segmented trajectories and compare them with the left-turn distance threshold. If the distance of the segmented trajectory is less than the left-turn distance threshold and the direction angle continuously and steadily increases, then it is considered a behavior of following a vehicle to turn left;

[0070] S4.6.6 Identify the end of the behavior of following a vehicle to turn left.

[0071] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0072] The present invention uses the shared bicycle GPS trajectory data of the traffic information public platform as the data source. Through the analysis of the shared bicycle GPS trajectory data, six shared bicycle dangerous behaviors such as dangerous speeding behavior, dangerous sudden deceleration behavior, dangerous parallel behavior, running a red light behavior, jaywalking behavior, and following a vehicle to turn left behavior are extracted. Based on the extracted index features, the recognition thresholds of six different dangerous driving behaviors are determined, and then a recognition model for dangerous speeding behavior, dangerous sudden deceleration behavior, dangerous parallel behavior, running a red light behavior, jaywalking behavior, and following a vehicle to turn left behavior based on thresholds is established, providing an effective, accurate, and scientific method for identifying shared bicycle dangerous behaviors.

[0073] The present invention improves the utilization rate of GPS data of shared bicycles. Through data preprocessing, the sparse shared bicycle data is densified, meeting the requirements of trajectory analysis, and accurately identifying dangerous driving behaviors such as speeding, running red lights, and sudden deceleration, which can effectively reduce the incidence of traffic accidents and ensure the safety of cyclists and other traffic participants; at the same time, through the analysis and monitoring of the behaviors of shared bicycles, it can help traffic management departments formulate more scientific management measures based on real data, improve the management efficiency of shared bicycles, guide cyclists to consciously abide by traffic rules, reduce uncivilized behaviors, and improve the overall traffic civilization level. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is a flowchart of a method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking;

[0075] Figure 2 is a flowchart of the data preprocessing of the present invention;

[0076] Figure 3 is a schematic diagram for judging dangerous speeding behavior of adjacent trajectory points within the same clustering set;

[0077] Figure 4 is a schematic diagram for judging dangerous sudden deceleration behavior of adjacent trajectory points within the same clustering set;

[0078] Figure 5 is a schematic diagram for judging dangerous parallel behavior of adjacent trajectory points within the same clustering set;

[0079] Figure 6 is a schematic diagram for running red lights behavior of adjacent trajectory points within the same clustering set;

[0080] Figure 7 is a schematic diagram for judging running red lights behavior of adjacent trajectory points within the same clustering set;

[0081] Figure 8 is a schematic diagram for judging the behavior of crossing the road;

[0082] Figure 9 is a schematic diagram for judging the behavior of following a vehicle to turn left;

[0083] Figure 10 is a flowchart for judging dangerous speeding behavior of adjacent trajectory points within the same clustering set;

[0084] Figure 11 is a flowchart for judging dangerous sudden deceleration behavior of adjacent trajectory points within the same clustering set;

[0085] Figure 12 is a flowchart for judging dangerous parallel behavior of adjacent trajectory points within the same clustering set;

[0086] Figure 13 Flow chart for judging red light running behavior of adjacent trajectory points within the same clustering set

[0087] Figure 14 Flow chart for judging jaywalking behavior

[0088] Figure 15 Flow chart for judging following a vehicle to turn left Specific implementation manners

[0089] The following further clarifies the present invention in conjunction with the accompanying drawings and specific implementation manners. It should be understood that the following specific implementation manners are only used to illustrate the present invention and not to limit the scope of the present invention.

[0090] A shared bicycle dangerous behavior identification method based on GPS trajectory data tracking according to the present invention analyzes and mines the GPS trajectory data of shared bicycle orders, extracts characteristic indicators of six dangerous driving behaviors including dangerous speeding behavior, dangerous sudden deceleration behavior, dangerous parallel behavior, red light running behavior, jaywalking behavior, and following a vehicle to turn left behavior, determines the recognition thresholds of the six different dangerous behaviors, and further establishes recognition models for identifying dangerous speeding riding, dangerous sudden deceleration riding, dangerous parallel behavior, red light running riding, jaywalking riding, and following a vehicle to turn left riding, so as to achieve accurate identification of these six dangerous behaviors.

[0091] Referring to the attached Figure 1 , the present invention defines six shared bicycle dangerous behaviors, extracts corresponding threshold indicators through historical data, and establishes judgment models for the six dangerous behaviors. After preprocessing the trajectory data set of shared bicycle orders, the road network data is imported to traverse the GPS trajectory data of all orders for road network matching, and the six dangerous behaviors are identified according to the recognition thresholds of six dangerous behaviors including dangerous speeding behavior, dangerous sudden deceleration behavior, dangerous parallel behavior, red light running behavior, jaywalking behavior, and following a vehicle to turn left behavior. When V n，t >V m , and when overtaking normal riding vehicles, or when the normal riding vehicle makes an avoidance behavior or the trajectory distance from the normal riding vehicle < 0.5 m, it is determined that a dangerous speeding behavior occurs; when TTC < 1 s, it is determined that a dangerous sudden deceleration behavior occurs; a red light waiting area is set. When the shared bicycle crosses the red light waiting area and continues to ride, its direction angle α n,t is compared. If α n,t <α m , it is determined that a red light running behavior occurs; when d nm,t <ρ n,t , and when the subsequent vehicle trajectory characteristics are similar to those of the parallel vehicle, it is determined that a dangerous parallel behavior occurs. When the trajectory point deviates from the original road and the direction angle of the segmented trajectory where the trajectory point is located Distance perpendicular to the road center line When it is, it is determined as a behavior of crossing the road; the trajectory within the crossroads area is separately segmented, and when the direction angles of the starting point and the ending point of the segmented trajectory It is recognized as a left-turn trajectory. If there are no stopping points among the trajectory points of the left-turn trajectory except the starting point and the ending point, the direction angle of the trajectory points is steadily increasing, and Then it is considered that a left-turn behavior has occurred.

[0092] A method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking, including the following steps:

[0093] S1 The data preprocessing steps required by the present invention include the following parts:

[0094] Traverse the GPS trajectory data set of shared bicycles, and screen abnormal trajectory points based on the satellite positioning signal strength threshold: when the signal strength is lower than the preset threshold, it is determined as low signal-to-noise ratio positioning data; further calculate the speed and displacement difference between adjacent trajectory points, and if the speed or displacement exceeds the preset range, it is determined as an abnormal point and removed.

[0095] Identify jump points according to the time stamp difference and spatial displacement of the trajectory points: if the time interval and displacement amount exceed the preset threshold, it is marked as a jump point. For the missing path between jump points, the dynamic programming algorithm is used to calculate the shortest path, and a weight function is constructed by combining the path length and time difference to generate complementary trajectory points to restore the path continuity.

[0096] Perform spline interpolation on the discrete trajectory points to generate a continuous trajectory sequence through a piecewise polynomial function. The interpolation interval is set according to the preset time step to ensure that the trajectory point density meets the requirements of subsequent analysis.

[0097] Convert the original GPS coordinates to the UTM coordinate system to eliminate projection errors, and at the same time standardize the time stamps, unify them into the reference time format and eliminate the time zone deviation. Divide the trajectory data based on the preset time window length, adopt the sliding window strategy to ensure data coverage integrity, and perform confidence verification on the number of trajectory points within the window.

[0098] Allocate time stamps to the interpolation points according to the average speed of the order trajectory, combined with the uniform motion assumption, to ensure that each trajectory point has a continuous and reasonable time mark.

[0099] Using the order start time as the index, group the trajectory data by the density-based clustering algorithm. During the clustering process, comprehensively evaluate the spatial distance, time stamp difference and core point density between trajectory points, and output a set of trajectory clusters with spatio-temporal similarity.

[0100] Integrate road network data and public geographic information data, map trajectory points to the road network through a probability model, and divide hot spots and non-hot spots based on the statistical results of the spatial density of trajectory points, supporting traffic planning and management decisions.

[0101] Finally, the data preprocessing is completed, and the processed GPS trajectory dataset of shared bicycles is obtained. As Figure 2 shown.

[0102] S2 classifies shared bicycle dangerous behaviors into six types: dangerous speeding behavior, dangerous sudden deceleration behavior, dangerous parallel behavior, running red lights behavior, crossing the road behavior, and following a vehicle to turn left behavior. And extract the threshold indicators for judging the six dangerous behaviors. The specific process is as follows:

[0103] Referring to local laws and regulations, classify dangerous behaviors into dangerous speeding behavior, dangerous sudden deceleration behavior, dangerous parallel behavior, running red lights behavior, crossing the road behavior, and these six dangerous behaviors, and then define the six dangerous behaviors respectively. The specific definitions are as follows:

[0104] Speeding behavior: Referring to the local traffic control regulations, the basic principle for identifying non-motor vehicles is to compare the instantaneous speed V n，t of the non-motor vehicle trajectory point with the speeding threshold V m . When the instantaneous speed V n，t of the non-motor vehicle trajectory point is greater than the speeding threshold V m , it is determined that the shared bicycle has committed a speeding behavior during this process. When the speeding vehicle passes by a normally riding vehicle, the lateral distance from the normally riding vehicle is less than the safety distance, and the safety distance is 0.5m, or the normally riding vehicle makes an avoidance behavior, then it is judged that the speeding vehicle has committed a dangerous speeding behavior, as Figure 3 shown.

[0105] Sudden deceleration behavior: Use the speeds V n，t of the front and rear trajectory points, the interval time T, and the distance d nm,t between adjacent trajectory points in the same clustering set during the deceleration state process to calculate the deceleration of each trajectory interval . When the deceleration is less than the deceleration threshold a m , it is determined that the shared bicycle has committed a dangerous sudden deceleration behavior during this process. Extract the normal trajectory following the sudden deceleration trajectory, and calculate the time to collision TTC nm,t between the normally riding vehicle and the suddenly decelerating vehicle during deceleration. If TTC nm,t <1s (including braking time and human reaction time), then it is considered that the suddenly decelerating vehicle has committed a dangerous sudden deceleration behavior, as Figure 4 shown.

[0106] Dangerous parallel behavior: Use adjacent trajectory points of different orders in the same clustering set to calculate the inter-point distance d between adjacent trajectory points n,j , Arbitrarily select one point among the adjacent trajectory points as the origin, and calculate the direction angle α of other trajectory points to this point n,t , Based on the direction angle α n,t Taking 0.5m as the semi-minor axis and 1.5m as the semi-major axis, calculate the polar radius ρ corresponding to this direction angle n,t , When the inter-point distance d n,j is less than the corresponding polar radius ρ n,t and there are continuously multiple inter-point distances d of adjacent trajectory points nm,t less than the corresponding polar radius ρ n,t it is determined that a parallel behavior has occurred in this process. If the difference between the parallel vehicle spacing and the road width is less than 1m at this time, it will cause other traffic participants to be unable to pass safely when there is an overtaking need, then it is considered that the parallel vehicle obstructs the traffic flow and is identified as a dangerous parallel behavior, and the process ends. As Figure 5 shown

[0107] Red light running behavior: Traverse the order trajectory data in each clustering set. By analyzing the density of trajectory points within a specific time interval, mark the area within 10 meters of the crosswalk coordinates and with high density as the red light waiting area. These areas are defined as geographical areas where trajectory points gather and stay for more than a preset threshold during the red light period. When a shared bicycle crosses the red light waiting area and continues to ride, compare its direction angle α n,t , If α n,t <α m , it is determined as a red light running behavior; as Figure 6 and Figure 7 shown

[0108] Jaywalking behavior: When the trajectory points deviate from the original road and there is a large change in the direction angle of the segmented trajectory where the trajectory points are located, the direction angle and the displacement of the starting point and ending point of the segmented trajectory on the coordinate axis perpendicular to the road i.e., half of the road width, it is determined as a jaywalking behavior; as Figure 8 shown

[0109] Following vehicle turning left behavior: Segment the trajectory within the crossroads area separately, and identify the direction angles of the starting point and ending point of the segmented trajectory as the left-turn trajectory. If there are no stopping points among the trajectory points of the left-turn trajectory except the starting point and ending point, the direction angle of the trajectory points is steadily increasing, and the length of the trajectory then it is considered that a following vehicle turning left behavior has occurred. As Figure 9 shown

[0110] S3 Determine the recognition thresholds for six types of abnormal driving behaviors.

[0111] Extract the characteristic indicators of six different dangerous behaviors from the GPS trajectory data of shared bike orders and determine the threshold indicators for the six different dangerous behaviors. The specific process is as follows:

[0112] Determine the relative speed threshold for dangerous speeding behavior:

[0113] Traverse all the shared bike order trajectory datasets, extract the trajectory point coordinates, the order start time start_time, and the order end time end_time. Calculate the instantaneous speed V of the shared bike trajectory points n，t , and sort them from smallest to largest. According to historical data analysis, extract the 80th percentile of the trajectory point speeds as the speeding threshold.

[0114] The parameters include: the coordinates (long n,t , lat n,t ) of the trajectory point at time t of the nth trajectory and the coordinates (long n,t+T , lat n,t+T ) of the trajectory point at time t + T of the nth trajectory, the time t when the trajectory point is adopted, the speed V of the shared bike n，t and the speeding threshold V m . The speed when a dangerous speeding behavior occurs is V n，t .

[0115]

[0116] Determine the threshold indicators for dangerous sudden deceleration while riding:

[0117] By traversing the speed V of all trajectory points n，t , and the time interval T between adjacent trajectory points, calculate the deceleration of each trajectory interval and sort them from largest to smallest. Select the 90th percentile as the deceleration threshold according to historical data According to the actual braking time and human reaction time, determine that the TTC nm,t threshold is 1s.

[0118] The parameters include: the speed V of the shared bike at the trajectory point at time T + t of the nth trajectory n，T+t ; the speed V of the shared bike at the trajectory point at time T of the nth trajectory n,t ; S n,t is the distance between adjacent trajectory points of the same trajectory, and the distance X between the deceleration starting point of deceleration trajectory n and the trajectory point of subsequent trajectory m with the same time stamp nm,t .

[0119] The deceleration when a dangerous sudden deceleration behavior occurs is

[0120]

[0121] The collision time threshold for emergency deceleration in case of danger is TTC nm,t .

[0122]

[0123] △V nm,t =V n,t -V m+m,t

[0124]

[0125] Determine the threshold index for dangerous parallel riding:

[0126] By performing DBSCAN clustering analysis on the data, a clustering set with the same time window, high trajectory similarity, and similar speeds is obtained. Traverse each clustering set and calculate the Euclidean distance d between the corresponding trajectory points of adjacent order trajectory data nm,t . Traverse the corresponding trajectory points of adjacent order trajectory data in each clustering set. Taking a certain point as the coordinate origin, calculate the direction angle α of the corresponding trajectory point n,t . Taking the current trajectory point as the origin, with a semi-minor axis of 0.5 meters and a semi-major axis of 1.5 meters, calculate the polar radius ρ of the direction angle of the corresponding trajectory point n,t , as the parallel threshold for this direction angle.

[0127] The parameters include: the semi-minor axis a of the ellipse = 0.5m, the semi-major axis b of the ellipse = 1.5m; the direction angle α of the parallel vehicle trajectory point n,t .

[0128] The threshold for dangerous parallel behavior of shared bicycles is ρ n,t .

[0129]

[0130] Determine the threshold index for running a red light while riding:

[0131] Based on historical data and local traffic regulations, confirm that the threshold index for running a red light is the direction angle α m .

[0132] The parameters include: the coordinates (long n,t , lat n,t ) of the trajectory point at time t of the nth trajectory. The coordinates (long n,t+T , lat n,t+T ) of the trajectory point at time t+T of the nth trajectory.

[0133] When a red light running behavior occurs, the direction angle of the trajectory point is α n,t .

[0134] α n,t= |atan2(x, y)|

[0135] y = sin(Δlong)·cos(lat n,t+T )

[0136] x = cos(lat n,t )sin(lat n,t+T ) - sin(lat n,t )cos(lat n,t+T )cos(Δlong)

[0137] Δlong = long n,t+T - long n,t

[0138] Determine the threshold index for jaywalking behavior:

[0139] According to historical data analysis, confirm that the threshold index for jaywalking is half of the road width

[0140] The parameters include: the direction angle α of the trajectory points of the segmented trajectory n,t , the coordinates (long n,t , lat n,t ) of the trajectory point of the nth trajectory at time t. The coordinates (long n,t+T , lat n,t+T ) of the trajectory point of the nth trajectory at time t + T.

[0141] Determine the threshold index for following a vehicle to turn left:

[0142] Extract the set of segmented trajectories passing through the intersection area, calculate the distance of the left-turn trajectories in the segmented trajectories that have stopping points in addition to the start and end points, and select the minimum distance as the left-turn distance threshold S min .

[0143] The parameters include: the trajectory point coordinates (long n,t , lat n,t ), the direction angle α of the trajectory point n,t , the crosswalk coordinates (long c , lat c ). The distance of the trajectory where the behavior of following a vehicle to turn left occurs is

[0144] S4 Traverse the GPS trajectory data of bike-sharing orders to identify dangerous behaviors.

[0145] Build a threshold-based dangerous behavior recognition model. By traversing the GPS trajectory data within different clustering sets, and based on the recognition thresholds of six dangerous behaviors such as dangerous speeding, dangerous sudden deceleration, dangerous parallel driving, and running red lights, identify these six dangerous cycling behaviors. The specific process is as follows:

[0146] Recognition of dangerous speeding behavior based on the GPS trajectory data of shared bike orders:

[0147] S4.1.1 Traverse all the shared bike trajectory data after data preprocessing, obtain the position changes and corresponding time intervals of all trajectory points, calculate the vehicle speeds of all trajectory points and take them as the instantaneous speeds of these trajectory points, and sort the speeds from small to large.

[0148] S4.1.2 Introduce POI data. Specifically, obtain the POI information of a specific area through a public data source, and extract the 80th percentile of the trajectory point speeds in the hot spot area during a specified time period as the speeding threshold for this area during this time period; in non-hot spot areas, also extract the 80th percentile of the trajectory speeds as the speeding threshold, so that the threshold decreases in places with high traffic flow and increases in places with low traffic flow, which is in line with the actual situation;

[0149] S4.1.3 When the vehicle speeds of consecutive trajectory points in the trajectory dataset exceed the relative speed threshold and this behavior lasts for more than 3 seconds, identify these shared bike driving behaviors as suspected dangerous speeding behaviors;

[0150] S4.1.4 Traverse the suspected dangerous speeding behaviors, and define that when the distance between the shared bike trajectory point and the road center line increases, that is, when deviating from the road center line, it is a cycling behavior of pulling over to avoid.

[0151] S4.1.5 Traverse the set of suspected dangerous speeding behaviors, calculate the Euclidean distance between the trajectory points of the suspected dangerous speeding behaviors and the normal behaviors at the same time stamp. When the trajectory points of the suspected dangerous speeding behaviors exceed the trajectory points of the normal behaviors, if the distance between the trajectory points of the suspected dangerous speeding behaviors and the normal behavior points is less than 0.5 meters, or the trajectory points of the normal behaviors have an avoidance behavior, then identify this suspected dangerous speeding behavior as a dangerous speeding behavior.

[0152] S4.1.6 Judge the end of the dangerous speeding behavior. See Figure 10 as shown.

[0153] Recognition of dangerous sudden deceleration behavior based on the GPS trajectory data of shared bike orders:

[0154] S4.2.1 Traverse the shared bike order trajectory dataset after data preprocessing, introduce POI data. Specifically, obtain the POI information of a specific area through a public data source, calculate the deceleration of the trajectory interval, and sort it from large to small;

[0155] S4.2.2 Extract the 90th percentile of the deceleration of the trajectory points in the hot spot area and non-hot spot area in each time period as the sharp deceleration threshold of this area in this time period;

[0156] S4.2.3 Traverse each trajectory in all clustering sets. When the deceleration of consecutive trajectory points in the trajectory is less than the sharp deceleration judgment threshold, it is identified as a suspected sharp deceleration behavior;

[0157] S4.2.4 Define the starting trajectory point where the deceleration exceeds the sharp deceleration judgment threshold within the trajectory interval. Traverse the set of suspected sharp deceleration trajectories, extract the trajectory points of the decelerated trajectories and the trajectory points of other normal driving trajectories in the clustering set, and calculate the vehicle distance Δd nm,t and the relative speed ΔV nm,t , and further calculate the time to collision, that is, Time To Collision. When TTC nm,t is less than 1 second, it is judged that the vehicle in front has a dangerous sharp deceleration behavior;

[0158] S4.2.5 Judge the end of the dangerous sharp deceleration behavior. See Figure 11 as shown.

[0159] Identification of dangerous parallel behavior based on GPS trajectory data of shared bicycle orders:

[0160] S4.3.1 Perform DBSCAN clustering analysis on the data to obtain clustering sets with the same time window, high trajectory similarity, and similar speeds, and verify the spatial position of the trajectories in combination with road network data;

[0161] S4.3.2 Traverse each clustering set, calculate the Euclidean distance d nm,t between the corresponding trajectory points of adjacent order trajectory data, and record the position of the trajectory points in the road network;

[0162] S4.3.3 Traverse the corresponding trajectory points of adjacent order trajectory data in each clustering set. Taking a certain point as the coordinate origin, calculate the direction angle α n,t of the corresponding trajectory points, and analyze the rationality of this direction angle in combination with road network data;

[0163] S4.3.4 Taking the current trajectory point as the origin, with a semi-minor axis of 0.5 meters and a semi-major axis of 1.5 meters, calculate the polar radius ρ n,t of the direction angle of the corresponding trajectory point as the parallel threshold of this direction angle, and record the parallel threshold at this direction angle to simplify the calculation;

[0164] S4.3.5 Traverse each cluster set, analyze the changes in coordinate positions and timestamps. If the Euclidean distance between the trajectory points corresponding to adjacent order trajectories is less than the parallel threshold, it is identified as a parallel behavior, and the rationality of this behavior is verified in combination with road network data;

[0165] S4.3.6 Calculate the lateral distance X of the vehicle with a dangerous parallel behavior in the direction perpendicular to the road n,t ; If the lateral distance X n,t and the difference between the width d of the road is less than 1 meter, and the subsequent other behaviors of the parallel vehicles are similar to the parallel trajectory within the same timestamp, it is determined that the leading vehicle has a dangerous parallel behavior;

[0166] S4.3.7 Determine the end of the dangerous parallel behavior. See Figure 12 as shown.

[0167] Red light running behavior recognition based on the GPS trajectory data of shared bike orders:

[0168] S4.4.1 Traverse the shared bike trajectory data, and use the DBSCAN clustering algorithm to divide the order trajectory data into multiple cluster sets according to the spatio-temporal characteristics of the trajectory and the similarity of the time window. During the clustering process, consider the distance between trajectory points, the timestamp difference, and the number of core points in each cluster.

[0169] S4.4.2 Import the road network data with crosswalk coordinates and match the road network data with the shared bike GPS trajectory data;

[0170] S4.4.3 Traverse the order trajectory data in each cluster set, and identify the red light waiting areas by analyzing the density of trajectory points within a specific time interval. These areas are defined as geographical areas where trajectory points gather and stay for more than a preset threshold during the red light period.

[0171] S4.4.4 For the trajectories outside the red light waiting areas, analyze the changes in their coordinate positions and direction angles. If the direction angle change is less than the set turning threshold without reasonable stops and does not match the traffic signal phase, it is determined as a red light running behavior. If it exceeds the set turning threshold, it is considered a turning behavior. Set the lower and upper limits of the turning threshold according to the actual situation as

[0172] S4.4.5 Identify the end of the red light running behavior. See Figure 13 as shown.

[0173] Recognition of jaywalking behavior based on the GPS trajectory data of shared bike orders:

[0174] S4.5.1 Import the non-motor vehicle road network data of the data collection location, traverse the GPS trajectory data of the shared bicycle orders, and perform road network matching; extract the trajectories with 3 or more trajectory points deviating from the original road and the coordinate change in the direction perpendicular to the road exceeding half of the road width to obtain the set of trajectories to be detected;

[0175] S4.5.2 Traverse the set of trajectories to be detected, calculate the direction angles of the trajectory points, and segment the offset trajectory points to obtain trajectory offset segments. Compare the direction angles of the trajectory points before and after the trajectory offset segments. According to historical data analysis, when the change range of the direction angle is within ±8°, the trajectory points are considered straight trajectory points. Extract five straight trajectory points before and after the trajectory offset segments respectively. When the number of trajectory points is insufficient, extract the actual number of straight trajectory points to obtain the set of segmented trajectories to be detected;

[0176] S4.5.3 Traverse the set of segmented trajectories to be detected, calculate the direction angles of the trajectory points and the displacements of the horizontal and vertical coordinates of the starting and ending points of the segmented trajectories. If there is a trajectory point whose direction angle change is greater than within the sampling time interval T, and the displacements of the horizontal and vertical coordinates of the starting and ending points of the segmented trajectories are greater than half of the average road width, and the straight-line distance between the first offset point of the trajectory and the coordinates of the zebra crossing > 10m, then determine that this trajectory is a behavior of crossing the road;

[0177] S4.5.4 The identification of the behavior of crossing the road ends. See Figure 14 as shown.

[0178] Recognition of the behavior of following a vehicle to turn left based on the GPS trajectory data of shared bicycle orders:

[0179] S4.6.1 Take the coordinates of the zebra crossing at the crossroads as the boundary, traverse the GPS trajectory data of all orders, and extract the segmented trajectories of the orders passing through the crossroads within the boundary of this crossroads to obtain the set of segmented trajectories;

[0180] S4.6.2 Traverse all the order trajectories in the set of trajectories to be judged. For each trajectory, calculate the direction angles of multiple key trajectory points on the trajectory, including the trajectory point just entering the crossroads, several intermediate points, and the trajectory point after leaving the crossroads; analyze the change trend of these direction angles. If the change amount of the direction angle after leaving the crossroads is greater than compared with when entering, then it is recognized as a left-turn trajectory, thus obtaining the set of left-turn trajectories;

[0181] S4.6.3 Traverse all the segmented trajectories, calculate the speeds of the trajectory points in the segmented trajectories. If there are stopping points in the trajectory, and the direction angles of the trajectory points remain stable before the stopping points and increase steadily after passing through the stopping points, then take it as a normal left-turn trajectory; arrange the distances of the normal left-turn trajectories from small to large;

[0182] S4.6.4 Take the minimum value as the left-turn distance threshold S min ;

[0183] S4.6.5 Traverse the distances of all the remaining segmented trajectories and compare them with the left-turn distance threshold. If the distance of a segmented trajectory is less than the left-turn distance threshold and the direction angle continuously and stably increases, it is considered a following vehicle left-turn behavior;

[0184] S4.6.6 Identify the end of the following vehicle left-turn behavior. See Figure 15 as shown.

[0185] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can still be made, and these improvements and refinements all fall within the protection scope of the claims of the present invention.

Claims

1. A method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking, characterized by: The steps include: S1 performs basic data preprocessing on the GPS trajectory data of the shared bicycles; S2 performs basic data preprocessing on the GPS trajectory data of shared bicycles to identify six types of dangerous behaviors of shared bicycles: dangerous speeding, dangerous sudden deceleration, dangerous parallel behavior, running red lights, crossing the road, and following vehicles to turn left; S3 determines the thresholds of six types of dangerous driving behaviors by analyzing historical data, including but not limited to relative speed threshold, red light running threshold, sudden deceleration threshold, collision time threshold, dangerous parallel riding threshold, direction angle threshold, and left turn distance threshold; S4 traverses the data of all shared bicycle orders and uses a predefined method to identify dangerous driving behaviors; Specifically, when the relative speed of a shared bicycle is greater than the relative speed threshold, and the lateral distance between the trajectory point of the bicycle and the adjacent trajectory point is less than 0.5m or the adjacent trajectory has an evasive behavior, it is judged as a dangerous speeding behavior; when the deceleration of a shared bicycle is less than the sudden deceleration threshold, it is identified as a deceleration trajectory, and the starting trajectory point in the trajectory interval where the deceleration exceeds the sudden deceleration judgment threshold is defined as the deceleration starting point. At the same timestamp, the collision time between the deceleration starting point and the adjacent trajectory point of the rear vehicle is calculated. When the collision time is less than 1 second and the distance between the vehicles is rapidly reduced to less than 1 meter in a short period of time, it is judged that the front vehicle has a dangerous sudden deceleration behavior; a red light waiting area is set. When a shared bicycle crosses the red light waiting area and continues to ride, its direction angle is compared. If the direction angle is less than the threshold, it is judged as a red light running behavior; when the trajectory is similar and the time is short, the front vehicle is judged to have a dangerous sudden deceleration behavior. If the Euclidean distance between two shared vehicles with the same window is less than the dangerous parallel riding threshold, it is identified as suspected dangerous parallel behavior. The difference between the lateral distance of dangerous parallel riding vehicles and the road width is less than 1m, and the subsequent behaviors of parallel vehicles with the same timestamp remain similar to the parallel trajectory within the same timestamp, the interval between trajectory points is less than 5 meters, and the speed changes are consistent when the parallel vehicles decelerate and the direction angle changes, then it is determined that the front vehicle has a dangerous parallel behavior; when the trajectory point deviates from the original road and the segmented trajectory where the trajectory point is located has a large direction angle change, the direction angle change exceeds the direction angle threshold, and the displacement of the start and end points of the segmented trajectory on the coordinate axis perpendicular to the road is greater than half of the road width, it is determined to be a jaywalking behavior; the trajectory within the intersection area is segmented separately, and the direction angles of the start and end points of the segmented trajectory are greater than Identify it as a left-turn trajectory. If there is no stopping point in the trajectory points of the left-turn trajectory except the starting point and the end point, the direction angle of the trajectory points is steadily increasing, and the length of the trajectory is less than the left-turn distance threshold, it is judged as the left-turn behavior of the following vehicle.

2. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: Data preprocessing also includes the following steps: S1 performs basic data preprocessing on the GPS trajectory data of the shared bicycles; S2 traverses the data and removes track points whose errors exceed a certain threshold due to weak GPS signals; S3 determines the jump point based on the time difference and position, and uses the Floyd algorithm to calculate the shortest weighted path between the jump points to fill in the points; S4 uses the spline interpolation method to interpolate the trajectory points, making all order trajectory points dense; S5 converts the coordinate system of the trajectory point coordinates into the UTM coordinate system, and normalizes the time involved in the data; S6 traverses the data, taking 30 seconds as the time window, and divides the shared bicycle data in the same time window; S7 calculates the average speed of each order data and allocates time to each track point according to the average speed to ensure that each track point has time; S8 uses the order start time as the index to traverse the shared bicycle trajectory data and uses the DBSCAN clustering algorithm to divide the order trajectory data into multiple cluster sets according to the spatiotemporal characteristics of the trajectory and the similarity of the time window. In the clustering process, the distance between trajectory points, the timestamp difference, and the number of core points in each cluster are considered; S9 imports road network data and POI data obtained from public data sources, performs road network matching, and extracts hotspot areas and non-hotspot areas based on trajectory density.

3. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: The method also includes: the speed of the shared bicycle track is V n，t : Where n is the order number; t is the sampling time of the first sampled trajectory point; T is the sampling time interval of adjacent trajectory points; where (long n,t ,lat n,t ) is the coordinate longitude and latitude of the track point at time t of the nth track, converted to radians; (long n,t+T ,lat n,t+T ) is the coordinate longitude and latitude of the nth track point at time t+T, converted to radians; S n,t is the straight-line distance between adjacent trajectory points on the same trajectory.

4. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: The method also includes decelerating Where V n，t V is the speed of the shared bicycle in the nth trajectory at the time point t; n，t+T S is the speed of the shared bicycle in the nth trajectory at the time point t+T; n,t is the distance between adjacent trajectory points.

5. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: The method for identifying red light running also includes: direction angle α n,t , angles are uniformly expressed in radians: a n,t =|atan2(x,y)|sin(Δlong)·cos(lat2)x =bas(lat) n,t )sin(lat n,t+T )-sin(lat n,t )basket(lat n,t+T )cos(Δlong)long =long n,t+T -long n,t Among them (long n,t ,lat n,t ) is the coordinate longitude and latitude of the track point at time t of the nth track; (long n,t+T ,lat n,t+T ) is the coordinate longitude and latitude of the trajectory point at time t+T of the nth trajectory.

6. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: The method also includes: n,t : b = 1.5m, which is the semi-major axis of the ellipse; α n,t is the direction angle, converted to radians.

7. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: The method also includes a collision time TTC nm,t : △V nm,t =V n,t -V n+m,t Where X nm,t is the distance between the deceleration starting point of deceleration trajectory n and the trajectory point of the subsequent trajectory m with the same timestamp, △V nm,t is the relative speed between the deceleration starting point of the deceleration trajectory n and the trajectory point of the subsequent trajectory m with the same timestamp; (long n,t ,lat n,t ) and (long n+m,t ,lat n+m,t ) are the same timestamps, i.e., the coordinates of the trajectory point of the decelerating vehicle n and the trajectory point of the subsequent trajectory m at time t.

8. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: The method also includes that the method for identifying red light running behavior includes the following steps: S3.1.1 Traverse the shared bicycle trajectory data and use the DBSCAN clustering algorithm to divide the order trajectory data into multiple cluster sets based on the spatiotemporal characteristics of the trajectory and the similarity of the time window; in the clustering process, consider the distance between trajectory points, timestamp differences, and the number of core points in each cluster; S3.1.2 Import the road network data with the coordinates of the pedestrian crossings and match the road network data with the GPS track data of the shared bicycles; S3.1.3 traverse the order trajectory data in each cluster set, and by analyzing the density of trajectory points in a specific time interval, mark the areas close to the crosswalk coordinates with a high density of 10 meters as red light waiting areas; these areas are defined as geographic areas where trajectory points gather and stay for more than a preset threshold during the red light period; S3.1.3 For the trajectory that exceeds the red light waiting area and continues to drive, analyze its coordinate position change and the direction angle change of the trajectory point; if the direction angle change of the trajectory point is less than the set turning threshold without reasonable stopping, it is judged as running a red light; if it exceeds the set turning threshold, it is considered as turning behavior, and the lower and upper limits of the turning threshold are set according to the actual situation. S3.1.4 Identify the end of the red light running behavior.

9. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: The method also includes that the method for identifying dangerous speeding behavior includes the following steps: S3.2.1 traverse all shared bicycle trajectory data after data preprocessing, obtain the position changes and corresponding time intervals of all trajectory points, calculate the speed of all trajectory points and use it as the instantaneous speed of the trajectory point, and sort the speeds from small to large; S3.2.2 introduces POI data, specifically, obtains POI information of a specific area through a public data source, and extracts the 80th percentile of the trajectory point speed in the hotspot area of ​​a specified time period as the speeding threshold of the area in the time period; in non-hotspot areas, the 80th percentile of the trajectory speed is also extracted as the speeding threshold, so that the threshold is lowered in places with heavy traffic, and the threshold is increased in places with light traffic, which is in line with reality; S3.2.3 When the speed of consecutive trajectory points in the trajectory data set exceeds the relative speed threshold and the duration of this behavior is greater than 3 seconds, the driving behavior of these shared bicycles is identified as suspected dangerous speeding behavior; S3.3.3 defines that when the distance between the shared bicycle trajectory point and the road centerline increases, that is, when it deviates from the road centerline, it is considered as a pull-over avoidance behavior; S3.3.4 Traverse the set of order trajectories of suspected dangerous speeding behaviors, calculate the Euclidean distance between the trajectory points of suspected dangerous speeding behaviors and normal behaviors at the same timestamp, and when the trajectory points of suspected dangerous speeding behaviors exceed the trajectory points of normal riding behaviors, if the distance between the trajectory points of suspected dangerous speeding behaviors and the normal riding behaviors is less than 0.5 meters, or the trajectory points of normal behaviors produce an evasive action, then the suspected dangerous speeding behavior is deemed to be a dangerous speeding behavior; S3.3.5 Determine that the dangerous speeding behavior has ended.

10. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1, characterized in that: The method also includes that the method for identifying dangerous rapid deceleration behavior includes the following steps: S3.3.1 traverse the shared bicycle order trajectory dataset after data preprocessing, introduce POI data, specifically obtain POI information of a specific area through a public data source, calculate the deceleration of the trajectory interval, and sort it from large to small; S3.3.2 and extract the 90th percentile of the deceleration of the trajectory points in the hotspot area and the non-hotspot area in each time period as the rapid deceleration threshold of the area in the time period; S3.3.3 Traverse each trajectory in all cluster sets, and when the deceleration of consecutive trajectory points in the trajectory is less than the sudden deceleration judgment threshold, it is identified as a suspected dangerous sudden deceleration behavior; S3.3.4 defines the starting trajectory point where the deceleration exceeds the sudden deceleration judgment threshold within the trajectory interval as the deceleration starting point, traverses the suspected sudden deceleration trajectory set, extracts the trajectory points of the deceleration trajectory and other normal driving trajectory points in the cluster set, and calculates the vehicle distance Δd between the normal trajectory point and the deceleration trajectory point with the same timestamp after the start of the deceleration starting point nm,t and relative velocity ΔV nm,t , further calculate the collision time, that is, Time To Collision, when TTC nm,t If the time is less than 1 second, it is judged that the vehicle ahead has dangerous rapid deceleration behavior; S3.3.5 Determine the end of the dangerous rapid deceleration behavior.

11. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1 is characterized in that: The method also includes that the method for identifying dangerous parallel behaviors includes the following steps: S3.4.1 Perform DBSCAN cluster analysis on the data to obtain clusters with the same time window, high trajectory similarity, and similar speed, and verify the spatial position of the trajectory in combination with the road network data; S3.4.2 Traverse each cluster set and calculate the Euclidean distance d between the corresponding trajectory points of adjacent order trajectory data nm,t , and record the location of the trajectory point in the road network; S3.4.3 Traverse the corresponding trajectory points of the adjacent order trajectory data of each cluster set, take a certain point as the coordinate origin, and calculate the direction angle α of the corresponding trajectory point n,t , and analyze the rationality of the direction angle in combination with the road network data; S3.4.4 Take the current trajectory point as the origin, 0.5 m as the semi-minor axis, and 1.5 m as the semi-major axis, and calculate the polar radius ρ of the direction angle of the corresponding trajectory point n,t , as the parallel threshold of this direction angle, and record the parallel threshold under this direction angle to simplify the calculation; S3.4.5 traverse each cluster set, analyze the changes in coordinate position and timestamp, and if the Euclidean distance between the trajectory points corresponding to the adjacent order trajectory data is less than the parallel threshold, identify it as parallel behavior, and verify the rationality of the behavior in combination with the road network data; S3.4.6 Calculate the lateral distance X perpendicular to the road direction of the vehicles that are engaging in dangerous parallel behavior. n,t ; If the horizontal distance X n,t If the difference with the width d of the road is less than 1 meter, and the subsequent other behaviors of the parallel vehicle remain similar to the parallel trajectory at the same timestamp, the leading vehicle is considered to have engaged in dangerous parallel behavior; S3.4.7 Determine the end of the dangerous parallel behavior.

12. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1, characterized in that: The method also includes that the method for identifying jaywalking behavior includes the following steps: S3.5.1 Import the non-motorized road network data of the data collection site, traverse the GPS track data of the shared bicycle orders, and perform road network matching; extract the tracks with 3 or more track points deviating from the original road and the coordinate change in the direction perpendicular to the road exceeding half of the road width, and obtain the track set to be detected; S3.5.2 traverse the set of trajectories to be detected, calculate the azimuth of the trajectory points, and segment the offset trajectory points to obtain the trajectory offset segments, compare the azimuths of the trajectory points before and after the trajectory point offset segment, and analyze the historical data to determine that the trajectory point is a straight trajectory point when the azimuth change is within ±8°, respectively extract five straight trajectory points before and after the trajectory point offset segment, and extract the actual number of straight trajectory points when the number of trajectory points is insufficient to obtain the set of segmented trajectories to be detected; S3.5.3 Traverse the set of segmented trajectories to be detected, calculate the direction angle of the trajectory point and the displacement of the horizontal and vertical coordinates of the starting and ending points of the segmented trajectory, and if there is a trajectory point with a direction angle change greater than If the displacement of the horizontal and vertical coordinates of the starting and ending points of the segmented trajectory is greater than half of the average width of the road, and the straight-line distance between the coordinates of the first offset point of the trajectory and the zebra crossing is greater than 10m, then the trajectory is judged as a jaywalking behavior; S3.5.4 traverses all segmented trajectory sets to be detected, records the shared bicycle orders in which jaywalking occurs, and identifies the end of jaywalking.

13. The method for identifying dangerous behaviors of shared bicycles based on GPS trajectory data tracking according to claim 1, characterized in that: The method further includes that the method for identifying the left turn behavior of the following vehicle includes the following steps: S3.6.1: Taking the coordinates of the zebra crossing at the intersection as the boundary, traverse the GPS trajectory data of all orders, extract the segmented trajectories of the orders passing through the intersection within the boundary of the intersection, and obtain the segmented trajectory set; S3.6.2: Traverse all order trajectories in the set of trajectories to be judged. For each trajectory, calculate the direction angles of multiple key trajectory points on the trajectory, including the trajectory point just entering the intersection, several points in the middle, and the trajectory point after leaving the intersection; by analyzing the changing trend of these direction angles, if the change in the direction angle after leaving the intersection is greater than that when entering, It is identified as a left-turn trajectory, thus obtaining a left-turn trajectory set; S3.6.3: Traverse all segmented trajectories and calculate the speed of the trajectory points in the segmented trajectories. If there is a stop point in the trajectory, and the direction angle of the trajectory point remains stable before the stop point, and the direction angle increases steadily after passing the stop point, then it is regarded as a normal left turn trajectory; arrange the distances of the normal left turn trajectory from small to large; S3.6.4: Take the minimum value as the left turn distance threshold S min Traverse the distances of all remaining segmented trajectories and compare them with the left-turn distance threshold. If the distance of the segmented trajectory is less than the left-turn distance threshold and the direction angle continues to increase steadily, it is considered to be a left-turn behavior of the following vehicle; S3.6.6 Determine that the following vehicle has completed its left turn.

Images