A lane-level turning information detection method and device for a flat intersection

Abstract

This invention discloses a method and apparatus for detecting lane-level turning information at at-grade intersections, comprising: preprocessing the road network, cleaning trajectory data and matching maps to connect trajectory points with their respective road segments; an intersection guidance zone identification step, considering the intersection trajectory distribution characteristics and vehicle straight-ahead stopping information at stop lines, to detect the intersection guidance zone; an intersection lane identification step, based on the intersection guidance zone identification results, extracting trajectories related to the guidance zone and performing multi-angle denoising, thereby inferring the guidance zone lanes; and an intersection lane-level turning information detection step, considering the intersection lane identification results, performing lateral and longitudinal statistics on the turning trajectories of each lane, and employing a supervised learning method to detect the intersection lane-level turning information. This invention is the first attempt to identify intersection lane-level turning information based on crowdsourced trajectories, making up for the lack of systematic research in lane-level turning information mining in existing works.

Description

Technical Field

[0001] This invention belongs to the field of fine navigation information mining, and in particular relates to a method and device for detecting lane-level turning information at planar intersections. Background Technology

[0002] With the development of the Internet and IoT technologies and the widespread adoption of location acquisition technologies, crowdsourced trajectory data—which is low-cost, rapidly updated, easy to collect, and contains semantic information about roads—has made it possible to acquire richer and more refined digital road information. To guide vehicles in an orderly manner and reduce traffic congestion, lane turning information is typically marked before intersections, either on the ground or in the air at the intersection guidance area. However, existing research on road information mining based on trajectory data mainly focuses on intersection identification and road network construction. While some scholars have explored lane information, this primarily involves the number of lanes and lane line identification. Therefore, considering the need for lane-changing guidance at intersections, marking turning information for each lane at intersections is of great significance for enriching road network navigation information and providing users with more refined navigation services.

[0003] Intersection turning information is crucial for constructing navigable road networks. Although numerous recognition methods have emerged based on crowdsourced trajectory data, most focus on lane turning, specifically categorized into clustering and map matching methods. Clustering is the most mainstream method for intersection turning recognition. It involves identifying the intersection location, probing the intersection's range, and then using a clustering algorithm to cluster sub-trajectories within the intersection range or intersection entry / exit points, thereby analyzing intersection turning rules. Map matching considers that road turning is related not only to intersections but also to the arcs of two connected roads, expressing intersection turning constraints in a "segment-segment" format. Specifically, it matches trajectories with intersection segments, analyzes segment-to-segment trajectory statistics, and identifies intersection turning rules based on unsupervised or supervised classification methods. While these clustering and map matching methods provide research ideas for intersection turning recognition, they cannot effectively eliminate noisy turns and do not address lane-level turning recognition. Due to limitations in trajectory data accuracy and noise levels, mining lane-level turning information remains a significant research challenge in road network construction, resulting in limited research. Summary of the Invention

[0004] In view of this, considering that traffic management departments often mark turning lanes in the guidance area of ​​intersections, this invention, based on taxi trajectory data, adopts the research approach of "intersection guidance area - intersection lane - intersection lane-level turning" to provide a lane-level turning information detection method for at-grade intersections, including the following steps:

[0005] Step 1, Data Preprocessing: First, perform topology checks on the road network data and quality cleaning on the trajectory data. Then, perform map matching on the road network and trajectory data to connect trajectory points with their respective road segments.

[0006] Step 2, Intersection guidance zone identification: Analyze the distribution characteristics of the trajectory at the intersection to identify the core area of ​​the intersection, initially determine the stop line position, then further determine the stop line based on the straight-ahead stopping information, and finally identify the guidance zone range based on this;

[0007] Step 3, Intersection Lane Recognition: Extract the guide zone trajectory based on the trajectory turning information, then filter the drift points of the extracted guide zone trajectory outside and inside the road, and finally perform guide zone lane recognition based on this.

[0008] Step 4, Intersection Lane-Level Turning Information Detection: Statistical analysis of the turning trajectories of each lane is performed, and feature design is carried out. Finally, a supervised learning method is used to identify the lane-level turning at the intersection.

[0009] Furthermore, the data preprocessing described in step 1 specifically includes:

[0010] Step 1.1: Perform a topology check on the road network data, including road intersection checks and pseudo-node checks;

[0011] Step 1.2 involves four basic preprocessing steps for the crowdsourced trajectory data: noise filtering, stop point detection, trajectory segmentation, and map matching.

[0012] Furthermore, the intersection guidance area identification in step 2 specifically includes:

[0013] Step 2.1: Determine the intersection core area based on trajectory distribution. Using the intersection as the center and K1 meters as the radius, construct a circular buffer zone. Divide the intersection space into K2 intervals from 0 to 360 degrees for trajectory distribution analysis. The maximum value of the difference index between the rings is determined as the radius of the intersection core area, thus defining the intersection core area. The specific calculation of the difference index Indexj_out between the rings is as follows:

[0014] Index j _ in=Std(num 1j , …, num ij , …, num 12j ) / mean(num 1j , …, num ij , …,num 12j (1)

[0015] Index j _out = Indexj+1 _in- Index j _in(2)

[0016] Where num ij For the first j The inner ring i Number of points in the interval trajectory; Std() is the standard deviation function; mean() is the mean function;

[0017] Step 2.2: Identify stop lines based on straight-ahead stopping trajectories. Starting from the nearest end of the intersection, construct a circular buffer zone along the opposite direction of the road segment, with a radius of K1 meters. Calculate the distribution density of straight-ahead stopping points within each circle: num / (πr). 2 -π(r-K1) 2 The intersection of the circular ring corresponding to the maximum stop point distribution density and the road segment is identified as the stop line position, where num is the number of straight stops in the circular ring and r is the radius of the circular buffer zone;

[0018] Step 2.3, the scope of the intersection guidance area is determined. The specific steps are as follows: (1) If the straight-ahead stopping point is less than or equal to C, the road segment cannot be more accurately detected based on the straight-ahead stopping point. Therefore, the intersection point of the road segment and the boundary of the core area of ​​the intersection is taken as the stop line position of the road segment; (2) For road segments with a straight-ahead stopping point greater than C and more stopping information, the stop line position is determined based on the straight-ahead stopping distribution. The specific implementation is the same as in step 2.2; (3) Based on the stop line, the [d, d+D] of the road segment is identified as the scope of the guidance area. The starting distance d is the distance from the stop line to the nearest road segment endpoint of the intersection, which is the upper cross section of the guidance area. The ending distance d+D is the lower cross section. D is a set constant in meters.

[0019] Furthermore, step 3, the intersection lane recognition, specifically includes:

[0020] Step 3.1: Extract the guide zone trajectory based on the identified intersection guide zone;

[0021] Step 3.2 After the guide area trajectory is extracted, the drift points of the trajectory outside the road and the drift points of the trajectory inside the road are further eliminated. The elimination of drift points of the trajectory outside the road is based on the Gaussian distribution of the trajectory in different directions of the two-way road. The trajectory points outside the intersection of the two-way trajectory distribution are regarded as noise and eliminated. The elimination of drift points of the trajectory inside the road is based on the maximum distance difference between the trajectory points in the trajectory and the road segment. The trajectory whose distance from the trajectory points in the trajectory to the road segment meets the formula (3) is judged as noise trajectory and eliminated.

[0022] |max (dis ( p 1, L ),…, dis ( p n ,L ))-min (dis ( p 1, L ),…, dis ( p n , L )) |>(W1-W2)(3)

[0023] in p k For the first in the trajectory k point; L For the section of road to enter; dis( p k , L ) is the first in the trajectory k Minimum distance from the point to the entry point, 1≤ k ≤ n ;max and min refer to the maximum and minimum operation respectively; W1 is the maximum design width of the intersection entrance lane; W2 is the maximum design width of the vehicle;

[0024] Step 3.3, lane identification in the guidance area: First, based on the road geometry features, road distribution features, and road topology connection features designed by the road network structure, as well as the road stopping information, one-way and two-way traffic, road width, road flow, and road speed designed based on the guidance area trajectory, machine learning algorithms are used to identify the number of lanes at the intersection. Then, based on the lane number information identification, a Gaussian mixture model is used to model the distance from the guidance area trajectory point to the road segment. The location of the intersection of adjacent Gaussian distributions is regarded as the boundary of adjacent lanes, i.e., the lane line position.

[0025] Furthermore, the specific implementation of step 3.1 is as follows: (1) Traverse the road segment to match the trajectory. If the trajectory is a straight trajectory, retain the intersection of the trajectory with the upper and lower cross sections of the guide area as the trajectory representation point. At the same time, determine whether there is a trajectory point that falls into the guide area. If there is, also retain the trajectory point that falls into the guide area. (2) If the trajectory is a left or right turning trajectory, determine whether there is a trajectory point that falls into the guide area. If there is, retain the trajectory point that falls into the guide area. If not, extend the starting turning point along the direction of the road segment. Retain the intersection of the extended line with the upper and lower cross sections as the trajectory representation point. (3) If the trajectory is a U-turn trajectory, directly extend the starting turning point along the direction of the road segment. Retain the intersection of the extended line with the upper and lower cross sections as the trajectory representation point.

[0026] Furthermore, the road geometric features include road segment length and curvature; the road distribution features include adjacent mesh area of ​​road segments, node distribution density of road segments, and road segment distribution density; and the road topological connectivity features include node clustering coefficient, degree centrality, proximity centrality, and betweenness centrality.

[0027] Furthermore, step 4, the detection of lane-level turning information at intersections, specifically includes:

[0028] Step 4.1, Lane Turning Trajectory Statistics. Specific statistical features include: lane turning trajectory count characteristics, lane turning trajectory lateral statistical characteristics, and lane turning trajectory longitudinal statistical characteristics, namely, the number of U-turns (CD), left turns (CL), straight-ahead turns (CS), and right turns (CR) in each lane; the percentage of U-turn trajectories in a lane relative to the total number of trajectories in that lane (HD), the percentage of left turns in a lane relative to the total number of trajectories in that lane (HL), the percentage of straight-ahead trajectories in a lane relative to the total number of trajectories in that lane (HS), and the percentage of right turns in a lane relative to the total number of trajectories in that lane (HR); the percentage of U-turn trajectories in a lane relative to the total number of U-turn trajectories in the guidance area (VD), the percentage of left turns in a lane relative to the total number of left turns in the guidance area (VL), the percentage of straight-ahead trajectories in a lane relative to the total number of straight-ahead trajectories in the guidance area (VS), and the percentage of right turns in a lane relative to the total number of right turns in the guidance area (VR).

[0029] Step 4.2, lane turning supervised recognition: randomly select P1% samples as training samples and P2% as test samples, based on the characteristics of lane turning trajectory statistics, i.e., the statistical number of lane turning trajectories. f c (CD, CL, CS, CR) Lateral statistical characteristics of lane turning trajectory f h (HD, HL, HS, HR) Longitudinal statistical characteristics of lane turning trajectory f v (VD, VL, VS, VR) uses machine learning algorithms to identify lane turning in the guidance area.

[0030] Furthermore, the machine learning algorithm is a random forest model.

[0031] Furthermore, K1 takes the value of 5, and K2 takes the value of 12.

[0032] The present invention also provides a lane-level turning information detection device for at-grade intersections, comprising the following modules:

[0033] The data preprocessing module is used to first perform topology checks on the road network data, clean the trajectory data, and then perform map matching of the road network and trajectory data to realize the connection between trajectory points and their respective road segments.

[0034] The intersection guidance zone identification module is used to analyze the distribution characteristics of the trajectory at the intersection to identify the core area of ​​the intersection, initially determine the stop line position, then further determine the stop line based on the straight-ahead stopping information, and finally identify the range of the guidance zone based on this.

[0035] The intersection lane recognition module is used to extract the guide zone trajectory based on the trajectory turning information, then filter the drift points of the extracted guide zone trajectory outside the road and inside the road, and finally perform guide zone lane recognition based on this.

[0036] The intersection lane-level turning information detection module is used to statistically analyze the turning trajectories of each lane and design features. Finally, a supervised learning method is used to identify the intersection lane-level turning.

[0037] Compared with existing technologies, the advantages and beneficial effects of this invention are as follows: Firstly, this invention focuses on the guidance area of ​​the entry segment, avoiding lane matching with the exit segment. Secondly, it considers the drift of the trajectory between lanes, which not only reduces error propagation and ensures the accuracy of the calculation results, but also effectively improves computational efficiency. This invention is the first to attempt to identify lane-level turning information at intersections based on crowdsourced trajectories, making up for the lack of systematic research in lane-level turning information mining in existing works. It can effectively solve the problems of long cycles and high costs in existing intersection fine navigation information mining. The mining scheme proposed in this invention is simple, efficient, and accurate, providing an effective technical reference for intersection fine navigation information mining. Attached Figure Description

[0038] Figure 1 This is a flowchart of a lane-level turning information detection method for at-grade intersections according to an embodiment of the present invention;

[0039] Figure 2 This is the intersection lane-level turning supervision classification and recognition process in this invention. Detailed Implementation

[0040] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0041] like Figure 1 As shown, the present invention provides a lane-level turning information detection method for at-grade intersections, comprising the following steps:

[0042] Step 1, Data Preprocessing: First, perform topology checks on the road network data and quality cleaning on the trajectory data. Then, perform map matching on the road network and trajectory data to connect trajectory points with their respective road segments.

[0043] Step 2, Intersection guidance zone identification: Analyze the distribution characteristics of the trajectory at the intersection to identify the core area of ​​the intersection, initially determine the stop line position, then further determine the stop line based on the straight-ahead stopping information, and finally identify the guidance zone range based on this;

[0044] Step 3, Intersection Lane Recognition: Extract the guide zone trajectory based on the trajectory turning information, then filter the drift points of the extracted guide zone trajectory outside and inside the road, and finally perform guide zone lane recognition based on this.

[0045] Step 4, Intersection Lane-Level Turning Information Detection: Statistical analysis of the turning trajectories of each lane is performed, and feature design is carried out. Finally, a supervised learning method is used to identify the lane-level turning at the intersection.

[0046] In this application, the data preprocessing step includes:

[0047] Step 1.1: Perform a topology check on the road network data, including road intersection checks and pseudo-node checks;

[0048] Step 1.2 involves four basic preprocessing steps for the crowdsourced trajectory data: noise filtering, stop point detection, trajectory segmentation, and map matching.

[0049] In this application, the intersection guidance area identification step includes:

[0050] Step 2.1: Determine the intersection core area based on trajectory distribution. Using the intersection as the center and a radius of 5 meters, construct a circular buffer zone. Divide the intersection space into 12 intervals from 0 to 360 degrees for trajectory distribution analysis. The maximum value of the difference index between the rings is determined as the radius of the intersection core area, thus defining the intersection core area. The specific calculation of the difference index Indexj_out between the rings is as follows:

[0051] Index j _ in=Std(num 1j , …, num ij , …, num 12j ) / mean(num 1j , …, num ij , …,num 12j (1)

[0052] Index j _out = Index j+1 _in- Index j _in(2)

[0053] Where num ij For the first j The inner ring i Number of points in the interval trajectory (1≤ j ≤20; 1≤ i ≤12); Std() is the standard deviation function; mean() is the mean function.

[0054] Step 2.2: Identify stop lines based on straight-ahead stopping trajectories. Starting from the nearest endpoint of the intersection, construct a circular buffer zone along the opposite direction of the road segment (unless otherwise specified, the road segment direction is consistent with the vehicle's travel direction), with a radius of 5 meters. Calculate the distribution density of straight-ahead stopping points within each circle: num / (πr). 2 -π(r-5) 2 The intersection of the circular ring and the road segment corresponding to the maximum stop point distribution density is identified as the stop line position, where num is the number of straight stops in the circular ring, r is the radius of the circular buffer zone, and 5≤r≤100.

[0055] Step 2.3, Determine the scope of the intersection guidance zone. The guide lane line area of ​​the road segment intersection cannot be accurately obtained. Therefore, according to the "Specifications for Setting Urban Road Traffic Signs and Markings" 4.8.3, which stipulates that the length of the guide lane line is generally not less than 30 meters, the 30-meter range before the stop line is identified as the guidance zone. The specific steps are as follows: (1) If the straight-ahead stopping point is less than or equal to C (default setting is 10), the road segment cannot be more accurately detected based on the straight-ahead stopping point. Therefore, the intersection of the road segment and the boundary of the core area of ​​the intersection is taken as the stop line position of the road segment; (2) For road segments with a straight-ahead stopping point greater than C and more stopping information, the stop line position is determined based on the straight-ahead stopping distribution, see step 2.2 for details; (3) Based on the stop line, identify [d, d+D] as the guidance zone range along the road segment. The starting distance d is the distance from the stop line to the nearest road segment endpoint of the intersection, which is the upper cross section of the guidance zone. The ending distance d+D is the lower cross section. D is set to 30 meters by default.

[0056] In this application, the intersection lane recognition step includes:

[0057] Step 3.1: Extract the guide zone trajectory based on the identified intersection guide zone. Specifically: (1) Traverse the road segment to match the trajectory. If the trajectory is a straight trajectory, retain the intersection of the trajectory with the upper and lower cross sections of the guide zone as the trajectory representation point. At the same time, determine whether there is a trajectory point that falls into the guide zone. If there is, retain the trajectory point that falls into the guide zone. (2) If the trajectory is a left or right turning trajectory, determine whether there is a trajectory point that falls into the guide zone. If there is, retain the trajectory point that falls into the guide zone. If not, extend the starting turning point along the direction of travel of the road segment and retain the intersection of the extended line with the upper and lower cross sections as the trajectory representation point. (3) If the trajectory is a U-turn trajectory, directly extend the starting turning point along the direction of travel of the road segment and retain the intersection of the extended line with the upper and lower cross sections as the trajectory representation point.

[0058] Step 3.2: After the guide area trajectory is extracted, further removal of drift points outside the road and drift points inside the road is performed. The removal of drift points outside the road is based on the Gaussian distribution of trajectories in different directions of the two-way road, and the trajectory points outside the intersection of the two-way trajectory distributions are regarded as noise and removed. The removal of drift points inside the road is based on the maximum distance difference between the trajectory points in the trajectory and the road segment, and the trajectory whose distance from the trajectory points in the trajectory to the entering road segment satisfies the formula (3) is judged as noise trajectory and removed.

[0059] |max (dis ( p 1, L ),…, dis ( p n , L ))-min (dis ( p 1, L ),…, dis ( p n , L )) |>(W1-W2)(3)

[0060] in p k For the first in the trajectory k point; L For the section of road to enter; dis( p k , L ) is the first in the trajectory k Minimum distance from the point to the entry point, 1≤ k ≤ n ;max and min refer to the maximum and minimum operating values, respectively; W1 is the maximum design width of the intersection entrance lane; W2 is the maximum design width of vehicles. According to the standard "Design Code for Urban Road Intersections", W1 can be set to 3.25, and according to the standard "Limits of External Dimensions, Axle Load and Mass of Road Vehicles", W2 can be set to 2.5.

[0061] Step 3.3, Lane Recognition in the Guiding Area. First, based on the road network structure, design road geometric features (segment length and curvature), road distribution features (segment adjacent mesh area, segment node distribution density, and segment distribution density), and road topological connectivity features (node ​​clustering coefficient, degree centrality, proximity centrality, and betweenness centrality). Based on the guiding area trajectory, design road stopping information, one-way / two-way traffic flow, road width, traffic volume, and road speed. Then, label training samples and use machine learning algorithms (such as the random forest model) to train the model. Based on the trained model, identify the number of lanes at intersections in the guiding area. Finally, based on the lane number information recognition, model the distance from the guiding area trajectory points to the road segments using a Gaussian mixture model. The location of the intersection of adjacent Gaussian distributions is considered the boundary of adjacent lanes, i.e., the lane line position.

[0062] In this application, the intersection lane-level turning information detection step includes:

[0063] Step 4.1, Lane Turning Trajectory Statistics. Specific statistical features include: the number of lane turning trajectories, lateral lane turning trajectory statistics, and longitudinal lane turning trajectory statistics, namely, the number of U-turn, left turn, straight, and right turn trajectories in each lane (CD, CL, CS, CR); the percentage of each lane's total number of trajectories (HD, HL, HS, HR); and the percentage of each lane's total number of U-turn, left turn, straight, and right turn trajectories within the guidance area (VD, VL, VS, VR).

[0064] Step 4.2, Lane Turning Supervision Recognition. 70% of the samples are randomly selected as training samples and 30% as test samples. Based on the statistical features of lane turning trajectories, namely the statistical features of the number of lane turning trajectories (CD, CL, CS, CR), the statistical features of the lateral direction of lane turning trajectories (HD, HL, HS, HR), and the statistical features of the longitudinal direction of lane turning trajectories (VD, VL, VS, VR), a machine learning algorithm (such as the random forest model) is used to identify lane turning in the guidance area.

[0065] The present invention also provides a lane-level turning information detection device for at-grade intersections, comprising the following modules:

[0066] The data preprocessing module is used to first perform topology checks on the road network data, clean the trajectory data, and then perform map matching of the road network and trajectory data to realize the connection between trajectory points and their respective road segments.

[0067] The intersection guidance zone identification module is used to analyze the distribution characteristics of the trajectory at the intersection to identify the core area of ​​the intersection, initially determine the stop line position, then further determine the stop line based on the straight-ahead stopping information, and finally identify the range of the guidance zone based on this.

[0068] The intersection lane recognition module is used to extract the guide zone trajectory based on the trajectory turning information, then filter the drift points of the extracted guide zone trajectory outside the road and inside the road, and finally perform guide zone lane recognition based on this.

[0069] The intersection lane-level turning information detection module is used to statistically analyze the turning trajectories of each lane and design features. Finally, a supervised learning method is used to identify the intersection lane-level turning.

[0070] The specific implementation methods and corresponding steps of each module are not described in this invention.

[0071] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for detecting lane-level turning information at at-grade intersections, characterized in that, Includes the following steps: Step 1, Data Preprocessing: First, perform topology checks on the road network data and quality cleaning on the trajectory data. Then, perform map matching on the road network and trajectory data to connect trajectory points with their respective road segments. Step 2, Intersection guidance zone identification: Analyze the distribution characteristics of the trajectory at the intersection to identify the core area of ​​the intersection, initially determine the stop line position, then further determine the stop line based on the straight-ahead stopping information, and finally identify the guidance zone range based on this; Step 2, the identification of the intersection guidance area, specifically includes: Step 2.1: Determine the intersection core area based on trajectory distribution. Using the intersection as the center and K1 meters as the radius, construct a circular buffer zone. Divide the intersection space into K2 intervals from 0 to 360 degrees for trajectory distribution analysis. The maximum value of the difference index between the rings is determined as the radius of the intersection core area, thus defining the intersection core area. The specific calculation of the difference index Indexj_out between the rings is as follows: Index j _ in=Std(num 1j , …, number ij , …, number 12j ) / mean(num 1j , …, number ij , …,number 12j )(1) Index j _out= Index j+1 _in- Index j _in(2) Where num ij For the first j The inner ring i Number of points in the interval trajectory; Std() is the standard deviation function; mean() is the mean function; Step 2.2: Identify stop lines based on straight-ahead stopping trajectories. Starting from the nearest road segment endpoint at the intersection, construct a circular buffer zone along the opposite direction of the road segment, with a radius of K1 meters. Calculate the distribution density of straight-ahead stopping points within each circle: num / (πr). 2 -π(r-K1) 2 The intersection of the circular ring corresponding to the maximum stop point distribution density and the road segment is identified as the stop line position, where num is the number of straight stops in the circular ring and r is the radius of the circular buffer zone; Step 2.3, the scope of the intersection guidance area is determined. The specific steps are as follows: (1) If the straight-ahead stopping point is less than or equal to C, the road segment cannot be more accurately detected based on the straight-ahead stopping point. Therefore, the intersection point of the road segment and the boundary of the core area of ​​the intersection is taken as the stop line position of the road segment; (2) For road segments with a straight-ahead stopping point greater than C and more stopping information, the stop line position is determined based on the straight-ahead stopping distribution. The specific implementation is the same as in step 2.2; (3) Based on the stop line, the [d, d+D] of the road segment is identified as the scope of the guidance area. The starting distance d is the distance from the stop line to the nearest road segment endpoint of the intersection, which is the upper cross section of the guidance area. The ending distance d+D is the lower cross section. D is a set constant in meters. Step 3, Intersection Lane Recognition: Extract the guide zone trajectory based on the trajectory turning information, then filter the drift points of the extracted guide zone trajectory outside and inside the road, and finally perform guide zone lane recognition based on this. Step 3, the intersection lane recognition, specifically includes: Step 3.1: Extract the guide zone trajectory based on the identified intersection guide zone; Step 3.2 After the guide area trajectory is extracted, the drift points of the trajectory outside the road and the drift points of the trajectory inside the road are further eliminated. The elimination of drift points of the trajectory outside the road is based on the Gaussian distribution of the trajectory in different directions of the two-way road. The trajectory points outside the intersection of the two-way trajectory distribution are regarded as noise and eliminated. The elimination of drift points of the trajectory inside the road is based on the maximum distance difference between the trajectory points in the trajectory and the road segment. The trajectory whose distance from the trajectory points in the trajectory to the road segment meets the formula (3) is judged as noise trajectory and eliminated. |max (dis ( p 1, L ) ,…, say ( p n , L ))-min (dis ( p 1, L ) ,…, say ( p n , L )) |>(W1-W2)(3) in p k For the first in the trajectory k point; L For the section of road to enter; dis( p k , L ) is the first in the trajectory k Minimum distance from the point to the entry point, 1≤ k ≤ n ;max and min refer to the maximum and minimum operation respectively; W1 is the maximum design width of the intersection entrance lane; W2 is the maximum design width of the vehicle; Step 3.3, lane identification in the guidance area: First, based on the road geometry features, road distribution features, and road topology connection features designed by the road network structure, as well as the road stopping information, one-way and two-way traffic, road width, road flow, and road speed designed based on the guidance area trajectory, machine learning algorithms are used to identify the number of lanes at the intersection. Then, based on the lane number information identification, a Gaussian mixture model is used to model the distance from the guidance area trajectory point to the road segment, and the location of the intersection of adjacent Gaussian distributions is regarded as the lane line position. Step 4, Intersection Lane-Level Turning Information Detection: Statistical analysis of the turning trajectories of each lane is performed, and feature design is carried out. Finally, a supervised learning method is used to identify the lane-level turning at the intersection.

2. The lane-level turning information detection method for at-grade intersections as described in claim 1, characterized in that: The data preprocessing described in step 1 specifically includes: Step 1.1: Perform a topology check on the road network data, including road intersection checks and pseudo-node checks; Step 1.2 involves four preprocessing steps for the crowdsourced trajectory data: noise filtering, stop point detection, trajectory segmentation, and map matching.

3. The lane-level turning information detection method for at-grade intersections as described in claim 1, characterized in that: The specific implementation method of step 3.1 is as follows: (1) Traverse the road segment to match the trajectory. If the trajectory is a straight trajectory, retain the intersection of the trajectory with the upper and lower cross lines of the guide area as the trajectory representation point. At the same time, determine whether there is a trajectory point that falls into the guide area. If there is, also retain the trajectory point that falls into the guide area. (2) If the trajectory is a left or right turning trajectory, determine whether there is a trajectory point that falls into the guide area. If there is, retain the trajectory point that falls into the guide area. If not, extend the starting turning point along the direction of the road segment. Retain the intersection of the extension line with the upper and lower cross lines as the trajectory representation point. (3) If the trajectory is a U-turn trajectory, directly extend the starting turning point along the direction of the road segment. Retain the intersection of the extension line with the upper and lower cross lines as the trajectory representation point.

4. The lane-level turning information detection method for at-grade intersections as described in claim 1, characterized in that: The road geometric features include road segment length and curvature; the road distribution features include adjacent mesh area of ​​road segments, node distribution density of road segments, and road segment distribution density; the road topological connectivity features include node clustering coefficient, degree centrality, proximity centrality, and betweenness centrality.

5. A lane-level turning information detection method for at-grade intersections as described in claim 1, characterized in that: Step 4, the intersection lane-level turning information detection, specifically includes: Step 4.1, Lane Turning Trajectory Statistics. Specific statistical features include: the number of U-turns (CD), left turns (CL), straight turns (CS), and right turns (CR) in each lane; the percentage of U-turn trajectories in a lane relative to the total number of trajectories in that lane (HD), the percentage of left turns in a lane relative to the total number of trajectories in that lane (HL), the percentage of straight turns in a lane relative to the total number of trajectories in that lane (HS), and the percentage of right turns in a lane relative to the total number of trajectories in that lane (HR); the percentage of U-turn trajectories in a lane relative to the total number of U-turn trajectories in the guidance area (VD), the percentage of left turns in a lane relative to the total number of left turns in the guidance area (VL), the percentage of straight turns in a lane relative to the total number of straight turns in the guidance area (VS), and the percentage of right turns in a lane relative to the total number of right turns in the guidance area (VR). Step 4.2, lane turning supervised recognition: randomly select P1% samples as training samples and P2% as test samples, based on the characteristics of lane turning trajectory statistics, i.e., the statistical number of lane turning trajectories. f c (CD, CL, CS, CR) Lateral statistical characteristics of lane turning trajectory f h (HD, HL, HS, HR) Longitudinal statistical characteristics of lane turning trajectory f v (VD, VL, VS, VR) uses machine learning algorithms to identify lane turning in the guidance area.

6. A lane-level turning information detection method for at-grade intersections as described in claim 5, characterized in that: The machine learning algorithm is the random forest model.

7. A method for detecting lane-level turning information at at-grade intersections as described in claim 1, characterized in that: The value of K1 is 5, and the value of K2 is 12.

8. A lane-level turning information detection device for at-grade intersections, implementing the lane-level turning information detection method for at-grade intersections as described in any one of claims 1-7, characterized in that, Includes the following modules: The data preprocessing module is used to first perform topology checks on the road network data, clean the trajectory data, and then perform map matching of the road network and trajectory data to realize the connection between trajectory points and their respective road segments. The intersection guidance zone identification module is used to analyze the distribution characteristics of the trajectory at the intersection to identify the core area of ​​the intersection, initially determine the stop line position, then further determine the stop line based on the straight-ahead stopping information, and finally identify the range of the guidance zone based on this. The intersection lane recognition module is used to extract the guide zone trajectory based on the trajectory turning information, then filter the drift points of the extracted guide zone trajectory outside the road and inside the road, and finally perform guide zone lane recognition based on this. The intersection lane-level turning information detection module is used to statistically analyze the turning trajectories of each lane and design features. Finally, a supervised learning method is used to identify the intersection lane-level turning.

Images