A method and system for detecting lane-level steering information based on unsupervised learning

Abstract

This invention discloses a lane-level turning information detection method and system based on unsupervised learning. It includes: a data preprocessing step, performing road network topology processing, trajectory data cleaning, and map matching to connect trajectory points with their respective road segments; an intersection lane information identification step, using a Gaussian mixture model to cluster matching trajectories within a certain range of adjacent endpoints at road segment intersections, thereby detecting lane information at road segment intersections; and a lane-level turning information mining step, considering different lane turning information and designing different lane turning information identification rules based on unsupervised classification methods. This invention conducts exploratory research on lane-level turning information identification in the direction of unsupervised classification, laying a certain technical foundation for refined navigation.

Description

Technical Field

[0001] This invention belongs to the field of navigation information mining, and in particular relates to a lane-level steering information detection method and system based on unsupervised learning. Background Technology

[0002] Lane turning markings at intersections are crucial for ensuring the orderly flow of large numbers of vehicles. They not only help traffic management departments control intersection traffic but also prevent drivers from taking detours due to choosing the wrong lane, providing key information for accurate navigation services. Existing research largely relies on high-resolution remote sensing, laser point cloud, and urban street view data for lane turning information recognition, which is costly and has a long update cycle. Summary of the Invention

[0003] To address the problems existing in the prior art, this invention has discovered that crowdsourced trajectory data is updated quickly and is easy to obtain, making it an important data source for road network identification. Therefore, this invention attempts to conduct unsupervised exploratory research on lane-level turning identification at intersections based on crowdsourced trajectory data without requiring extensive annotation, in order to lay the foundation for the mining of refined navigation information.

[0004] In view of this, the present invention provides a lane-level steering information detection method based on unsupervised learning, comprising 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 lane information identification: First, the matching trajectory of the road segment within a few meters of the nearest endpoint of the intersection of the study road segment is denoised using a delanuary triangulation. Then, based on the Gaussian mixture model, cluster analysis is performed on the minimum distance from the matching trajectory point to the road segment to obtain the number of lanes and lane line information of the intersection of the study road segment.

[0007] Step 3, Lane-level steering information mining: First, obtain lane steering trajectory information, then perform statistical analysis on the lane steering trajectory, and finally, based on different lane trajectory distributions, detect lane-level steering information according to different rules.

[0008] Furthermore, the specific implementation methods in step 3 include:

[0009] Step 3.1, Lane turning trajectory statistics: Count the number of U-turn, left turn, straight and right turn trajectories in each lane of the intersection, and then perform lateral and longitudinal statistics based on the number of turning trajectories in each lane;

[0010] Step 3.2, Unsupervised lane turning identification: Noisy turns are eliminated, and then corresponding rules are designed based on the lateral and longitudinal statistical information of the remaining turns to detect lane turning information.

[0011] Furthermore, step 3.1 involves lateral statistics, specifically counting the number C of U-turn trajectories within the lane. D Number of left-turn trajectories C L Number of straight trajectories C S Number of right-turn trajectories C R The percentage of the total number of trajectories in this lane is specifically represented by H. D H L H S H R The specific calculation formula is as follows:

[0012]

[0013] Where X∈(D,L,S,R); X lanei Let X be the number of turning trajectories in lane i.

[0014] Furthermore, step 3.1 involves longitudinal statistics, specifically counting the number C of U-turn trajectories in the lane. D The percentage of total U-turn trajectories and the number of left-turn trajectories in the guidance area (C) L The percentage of total left-turn trajectories and the number of straight-through trajectories in the guidance area (C) S The percentage of straight-ahead trajectories and the percentage of right-turn trajectories in the guidance area (C) R The percentage of the total number of right-turn trajectories in the guidance area is specifically represented by V. D V L V S V R The specific calculation formula is as follows:

[0015]

[0016] Where X∈(D,L,S,R); X lanei Let X be the number of turning trajectories in lane i, and n be the number of lanes.

[0017] Furthermore, the specific implementation method of step 3.2 is as follows:

[0018] Step 3.2.1, noise removal: Turns with a horizontal statistical percentage of less than p% or a vertical statistical percentage of less than p% are considered noise turns and removed.

[0019] Step 3.2.2, U-turn steering judgment: separate U-turns from other steering and perform lane-level steering information detection;

[0020] Step 3.2.3, Other steering discrimination: Different rules are designed for the steering trajectory distribution of the leftmost, rightmost and middle lanes to identify lane steering information.

[0021] Furthermore, the specific implementation method of step 3.2.2 is as follows:

[0022] Rule 1: Count the total number of U-turn trajectories on a road segment. If the total number is less than K1, then U-turns are not allowed in any lane of that road segment. If the total number is greater than K1, then U-turns are allowed on that road segment.

[0023] Rule 2: If a U-turn is permitted on a road segment, and after noise-induced turn-by-turn elimination is performed on all lanes of that segment, there are no remaining U-turn lanes, then the leftmost lane is assumed to be permitted for U-turns.

[0024] Rule 3: If a lane allows U-turns, and after noise-driven turns are eliminated, the remaining turns in that lane are U-turn turns, and that turn has the largest proportion, then that lane can only allow U-turns; otherwise, that lane can allow U-turns.

[0025] Furthermore, the specific implementation method of step 3.2.3 is as follows:

[0026] (1) If there is only one statistical direction X left in the statistics of left turn, straight, and right turn, then the direction of the lane must be X turn;

[0027] (2) If there are no remaining statistical directions in the statistics, then the two largest directions X and Y in the horizontal statistics of left turn, straight, and right turn before noise removal are used for discrimination;

[0028] (3) If the number of remaining statistical turns is greater than or equal to 2, then the two turns X and Y with the largest remaining lateral statistics are used for discrimination.

[0029] Furthermore, the specific implementation of the judgment based on the two largest lateral turning points X and Y in steps (2) and (3) is as follows, where L is left turning, R is right turning, and S is straight-ahead:

[0030] (1) Rule for determining the leftmost lane:

[0031] Rule 1: If there is an L-turn in either X or Y turns, then that lane has an L-turn lane;

[0032] Rule 2: If there is an R turn in the X and Y turns, and the longitudinal direction accounts for the largest proportion, then there is an R turn in that lane;

[0033] Rule 3: If there is an S turn in X or Y turns, determine if the lateral statistical proportion of the S turn is the smallest. If it is, the S turn is a noise turn. If not, determine if the longitudinal statistical proportion of the S turn is greater than 1 / k, where k is the number of lanes in the guidance area. If it is, then there is an S turn in that lane.

[0034] (2) Middle Lane Judgment Rules

[0035] Rule 4: If an L-turn exists, determine whether the longitudinal statistical proportion of L-turns is greater than 1 / k. If it is, then there is an L-turn in that lane; otherwise, it is a noise turn and is removed.

[0036] If an R-turn exists, determine whether the longitudinal statistical proportion of R-turns is greater than 1 / k. If it is, then the lane has an R-turn; otherwise, it is a noise turn and is removed.

[0037] Rule 5: If max(H) X H Y ) / (H X +H Y If the value of max(H) is greater than K3, then the lane turn will be identified as the turn with the largest lateral statistical proportion. X H Y ) / (H X +H Y If H is greater than K2 and less than K3, then the lane has both X and Y directions; otherwise, the lane has the lane with the smallest lateral proportion. x H y These refer to the lateral statistics for X-turn and Y-turn, respectively.

[0038] (3) Rules for determining the rightmost lane

[0039] Rule 6: If there is an R turn in either X or Y turn, then that lane has an R turn lane;

[0040] Rule 7: If there is an L-turn in both X and Y turns, and the longitudinal direction accounts for the largest proportion, then there is an L-turn in that lane;

[0041] Rule 8: If there is an S-turn in X or Y turns, determine if the lateral statistical proportion of the S-turn is the smallest. If so, the S-turn is a noise turn. If not, determine if the longitudinal statistical proportion of the S-turn is greater than 1 / k. If so, the lane has an S-turn.

[0042] This invention also provides a lane-level steering information detection system based on unsupervised learning, comprising the following modules:

[0043] The preprocessing module is used for data preprocessing: first, it performs topology checks on the road network data and quality cleaning on the trajectory data; then, it performs map matching of the road network and trajectory data to connect the trajectory points with their respective road segments.

[0044] The lane information recognition module is used for lane information recognition at intersections: First, the matching trajectory of the road segment within a few meters of the nearest endpoint of the intersection of the study road segment is denoised using a delanuary triangulation. Then, based on the Gaussian mixture model, cluster analysis is performed on the minimum distance from the matching trajectory point of the road segment to the road segment to obtain the number of lanes and lane line information of the intersection of the study road segment.

[0045] The steering information detection module is used for lane-level steering information mining: first, it acquires lane steering trajectory information, then performs statistical analysis on the lane steering trajectory, and finally, based on different lane trajectory distributions, it detects lane-level steering information according to different rules.

[0046] Compared with existing technologies, the advantages and beneficial effects of this invention are as follows: This invention takes into account the different distributions of turning trajectories in different lanes and employs an unsupervised classification method for designing rules for different lane turning information. This effectively avoids the negative impacts of supervised classification, such as the large number of sample annotations and the potential presence of erroneous, insufficiently representative, or incompletely covered samples. It compensates for the shortcomings of unsupervised classification research in lane-level turning information mining and provides effective technical guidance for lane-level turning information mining. Attached Figure Description

[0047] Figure 1 This is a flowchart of a lane-level steering information detection method based on unsupervised learning, according to an embodiment of the present invention.

[0048] Figure 2 This is a flowchart of the lane-level steering unsupervised classification and recognition process in this invention.

[0049] Figure 3 This is the experimental data and the distribution of experimental intersections in the embodiments of the present invention. Detailed Implementation

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

[0051] like Figure 1 As shown, the present invention provides a lane-level steering information detection method based on unsupervised learning, comprising the following steps:

[0052] 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.

[0053] Step 2, Intersection Lane Information Identification: First, the matching trajectory of the road segment within 30 meters of the nearest endpoint of the intersection of the study road segment is denoised using a Delanuary triangulation. Then, based on the Gaussian mixture model, cluster analysis is performed on the minimum distance from the matching trajectory point to the road segment to obtain the number of lanes and lane line information of the intersection of the study road segment.

[0054] Step 3, Lane-level steering information mining: First, obtain lane steering trajectory information, then perform statistical analysis on the lane steering trajectory, and finally design different rules to detect lane-level steering information based on different lane trajectory distributions.

[0055] Furthermore, the lane-level steering information mining described in step 3 specifically includes:

[0056] Step 3.1, Lane turning trajectory statistics: Count the number of U-turn (D), left turn (L), straight (S), and right turn (R) trajectories for each lane at the intersection, and then perform lateral and longitudinal statistics based on the number of turning trajectories for each lane.

[0057] Step 3.2, Unsupervised lane turning identification: Based on experience, turns with a lateral statistical percentage of less than 10% or a longitudinal statistical percentage of less than 10% are considered noise turns and are eliminated. Then, based on the lateral and longitudinal statistical information of the remaining turns, corresponding rules are designed to detect lane turning information.

[0058] Furthermore, the specific implementation method of step 3.1 is as follows:

[0059] Step 3.1.1, Lateral Statistics: This involves counting the number of U-turn trajectories in the lane (C D ), number of left-turn trajectories (C) L ), number of straight trajectories (C) S ), Number of right-turn trajectories (C) R The percentage of the total number of trajectories in this lane, specifically expressed as H. D H L H S H R The specific calculation formula is as follows:

[0060]

[0061] Where X∈(D,L,S,R); X lanei Let X be the number of turning trajectories in lane i, where 1 ≤ i ≤ n.

[0062] Step 3.1.2, Longitudinal Statistics: This involves counting the number of U-turn trajectories in the lane (C). D The percentage of total U-turn trajectories and the percentage of left-turn trajectories in the guidance area (C) L The percentage of left-turn trajectories and the percentage of straight-through trajectories in the guidance area (C)S The percentage of straight-ahead trajectories and the percentage of right-turn trajectories in the guidance area (C) R The percentage of the total number of right-turn trajectories in the guidance area is specifically represented by V. D V L V S V R The specific calculation formula is as follows:

[0063]

[0064] Where X∈(D,L,S,R); X lanei Let X be the number of turning trajectories in lane i, where 1 ≤ i ≤ n.

[0065] Furthermore, the specific implementation method of step 3.2 is as follows:

[0066] Step 3.2.1, noise removal: Turns with a horizontal statistical percentage of less than 10% or a vertical statistical percentage of less than 10% are considered noise turns and removed.

[0067] Step 3.2.2, U-turn detection: Because some areas lack U-turn directional arrows and no-U-turn signs, some lanes may exhibit U-turn trajectories, affecting the identification of other potential turns. Furthermore, U-turn trajectories are less frequent than other turn trajectories; therefore, this paper processes U-turns separately from other turns for lane-level turn information detection.

[0068] Step 3.2.3, Other turning determination: Since the turning trajectory distribution of the leftmost and rightmost lanes is different from that of the middle lane, we designed different rules to identify lane turning information.

[0069] Furthermore, the specific implementation method of step 3.2.2 is as follows:

[0070] Rule 1: Count the total number of U-turn trajectories on a road segment. If the total number is less than K1 (this parameter is set through parameter tuning, generally greater than 10), then U-turns are not allowed in any lane of that road segment. If the total number is greater than K1, then U-turns are allowed on that road segment.

[0071] Rule 2: If a U-turn is permitted on a road segment, and after noise-induced turn-by-turn elimination is performed on all lanes of that segment, there are no remaining U-turn lanes, then the leftmost lane is assumed to be permitted for a U-turn.

[0072] Rule 3: If a lane allows U-turns, and after noise-driven turns are eliminated, the remaining turns in that lane are U-turn turns, and that turn has the largest proportion, then that lane can only allow U-turns; otherwise, that lane can allow U-turns.

[0073] Furthermore, the specific implementation method of step 3.2.3 is as follows:

[0074] (1) If there is only one statistical direction X among the statistics of left turn, straight, and right turn, then the direction of the lane must be X turn.

[0075] (2) If there are no remaining statistical directions in the statistics, then the two largest directions X and Y in the horizontal statistics of left turn, straight and right turn before noise removal are used for discrimination.

[0076] (3) If the number of remaining statistical turns is greater than or equal to 2, then the two turns X and Y with the largest remaining lateral statistics are used for discrimination.

[0077] Furthermore, the specific implementation of steps (2) and (3) involving the discrimination based on the two largest lateral steering values ​​X and Y is as follows, where L steering is left steering, R steering is right steering, and S steering is straight-ahead:

[0078] (1) Rule for determining the leftmost lane:

[0079] Rule 1: If there is an L turn in the X or Y turn, then the lane has an L turn lane.

[0080] Rule 2: If there is an R turn in the X and Y turns, and the longitudinal direction accounts for the largest proportion, then there is an R turn in that lane.

[0081] Rule 3: If there is an S-turn in X or Y turns, determine if the lateral statistical proportion of the S-turn is the smallest. If so, the S-turn is a noise turn. If not, determine if the longitudinal statistical proportion of the S-turn is greater than 1 / k (k is the number of lanes in the guidance area). If so, the lane has an S-turn.

[0082] (2) Middle Lane Judgment Rules

[0083] Rule 4: If an L-turn exists, determine whether the longitudinal statistical proportion of L-turns is greater than 1 / k (k is the number of lanes in the road segment). If it is, then the lane has an L-turn; otherwise, it is a noisy turn and is removed.

[0084] If an R-turn exists, determine whether the longitudinal statistical proportion of R-turns is greater than 1 / k (k is the number of lanes in the road segment). If it is, then the lane has an R-turn; otherwise, it is a noise turn and is removed.

[0085] Rule 5: If max(H) X H Y ) / (H X +H Y If the value of max(H) is greater than K3, then the lane turn is identified as the turn with the largest lateral statistical proportion. X H Y ) / (H X +H YIf H is greater than K2 and less than K3, then the lane has both X and Y directions; otherwise, the lane is identified as the direction with the smallest lateral proportion. x H y The X-turn lateral statistics and Y-turn lateral statistics refer to the statistics respectively, both of which are calculated using formula (1). The parameters K2 and K3 are also set through parameter tuning. According to common sense, K2 can generally be set to 0.4 or 0.5, and K3 is generally greater than or equal to 0.5.

[0086] (3) Rules for determining the rightmost lane

[0087] Rule 6: If there is an R turn in the X or Y turn, then the lane has an R turn lane.

[0088] Rule 7: If there is an L-turn in the X and Y turns, and the longitudinal direction accounts for the largest proportion, then there is an L-turn in that lane.

[0089] Rule 8: If there is an S turn in X or Y turns, determine whether the lateral statistical proportion of the S turn is the smallest. If so, the S turn is a noise turn. If not, determine whether the longitudinal statistical proportion of the S turn is greater than 1 / k (k is the number of lanes in the road segment). If so, the lane has an S turn.

[0090] In practical implementation, the method proposed in this invention can be automatically executed by those skilled in the art using computer software technology. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of this invention and computer equipment including the computer program running the corresponding computer program, should also be within the scope of protection of this invention. This invention also provides a lane-level steering information detection system based on unsupervised learning, comprising the following modules:

[0091] The preprocessing module is used for data preprocessing: first, it performs topology checks on the road network data and quality cleaning on the trajectory data; then, it performs map matching of the road network and trajectory data to connect the trajectory points with their respective road segments.

[0092] The lane information recognition module is used for lane information recognition at intersections: First, the matching trajectory of the road segment within a few meters of the nearest endpoint of the intersection of the study road segment is denoised using a delanuary triangulation. Then, based on the Gaussian mixture model, cluster analysis is performed on the minimum distance from the matching trajectory point of the road segment to the road segment to obtain the number of lanes and lane line information of the intersection of the study road segment.

[0093] The steering information detection module is used for lane-level steering information mining: first, it acquires lane steering trajectory information, then performs statistical analysis on the lane steering trajectory, and finally, based on different lane trajectory distributions, it detects lane-level steering information according to different rules.

[0094] The specific implementation methods of each module are the same as those of each step, and will not be described in this invention.

[0095] To verify the effectiveness of the method, two areas within a certain region were selected for experimentation. Regarding the basic road network, an OSM road network (such as...) was adopted. Figure 3 (a) and (c) in the data; regarding trajectory data, the crowdsourced trajectories of January 21, 2021 were used (e.g., ...). Figure 3 Experiments were conducted using (b) and (d) in the model. The trajectory point sampling interval was mostly concentrated within 0-10 seconds, with a positioning accuracy of 0-5 meters. Ten representative intersections located on main urban roads were selected as the research subjects (e.g., Figure 3 (a) and (c) in red boxes) include not only the cross intersections, T-junctions and five-way intersections involved in at-grade intersections, but also simple elevated intersections.

[0096] To verify the rationality of the unsupervised classification method, experiments were conducted at 10 intersections. Each of the 10 intersections corresponded to 3-5 guide zones, and each guide zone corresponded to a certain number of lanes, totaling 183 lanes. We tuned the parameters K1, K2, and K3 through repeated experiments. When K1, K2, and K3 were set to 20, 0.5, and 0.5 respectively, the lane turning recognition results at the intersections were optimal, with an overall recognition accuracy of 77.0%.

[0097] To validate the unsupervised lane turning classification method, experiments were conducted using threshold segmentation for lane-level turning recognition without distinguishing between the leftmost, rightmost, and middle lanes. The overall recognition accuracy for all 10 intersections was only 66.1%, significantly lower than the proposed method, indicating that indiscriminate threshold segmentation significantly reduces the overall accuracy. This demonstrates the necessity of considering the leftmost, rightmost, and middle lanes when conducting research on lane-level turning recognition.

[0098] To further verify the necessity of performing lateral and longitudinal statistics on lane turning trajectories to eliminate noise-induced turning, we detected turning information based on the designed turning recognition rules without eliminating noise-induced turning. The overall recognition accuracy rate for 10 intersections was 76.0%, indicating that the overall recognition rate was low without eliminating noise-induced turning.

[0099] To further evaluate the recognition results of the unsupervised classification method, the lane turning accuracy of each intersection was also statistically analyzed (as shown in Table 1). It can be seen that the lane turning accuracy of the threshold segmentation method is significantly lower than that of the unsupervised classification method. Furthermore, the lane turning accuracy of some intersections where noisy turns are not removed is also lower than that of the unsupervised classification method.

[0100] Table 1. Lane turning recognition results for each intersection using the unsupervised classification method.

[0101]

[0102]

[0103] 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 lane-level steering information detection method based on unsupervised learning, 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 lane information identification: First, the matching trajectory of the road segment within a few meters of the nearest endpoint of the intersection of the study road segment is denoised using a delanuary triangulation. Then, based on the Gaussian mixture model, cluster analysis is performed on the minimum distance from the matching trajectory point to the road segment to obtain the number of lanes and lane line information of the intersection of the study road segment. Step 3, Lane-level turning information mining: First, obtain lane turning trajectory information, then perform statistical analysis on the lane turning trajectory, and finally, based on different lane trajectory distributions, detect lane-level turning information according to different rules. The specific implementation methods in step 3 include: Step 3.1, Lane turning trajectory statistics: Count the number of U-turn (D), left turn (L), straight (S), and right turn (R) trajectories for each lane at the intersection, and then perform lateral and longitudinal statistics based on the number of each lane's turning trajectory. Step 3.2, Unsupervised lane turning identification: Noisy turns are eliminated, and then corresponding rules are designed based on the lateral and longitudinal statistical information of the remaining turns to detect lane turning information; The specific implementation method of step 3.2 is as follows: Step 3.2.1, noise removal: Turns with a horizontal statistical percentage of less than p% or a vertical statistical percentage of less than p% are considered noise turns and removed. Step 3.2.2, U-turn steering judgment: separate U-turns from other steering and perform lane-level steering information detection; Step 3.2.3, Other steering discrimination: Different rules are designed for the steering trajectory distribution of the leftmost, rightmost and middle lanes to identify lane steering information; The specific implementation method of step 3.2.3 is as follows: (1) If only one of the left turn, straight, and right turn statistics remains, then... X Then the lane must be turning. X Turning; (2) If there are no remaining statistical turns in the statistics, then the two turns X and Y with the largest values ​​in the horizontal statistics of left turn, straight, and right turn before noise removal are used to make the judgment; (3) If the number of remaining statistical turns is greater than or equal to 2, then the two turns X and Y with the largest remaining lateral statistics are used for discrimination; The specific implementation of the judgment based on the two largest lateral turns X and Y in steps (2) and (3) is as follows, where L is left turn, R is right turn, and S is straight turn: (21) Rule for determining the leftmost lane: Rule 1: If X , Y There is a turning point L If you turn, then the lane exists. L Turning lane; Rule 2: If X , Y There is a turning point R If the lane is turning and has the largest longitudinal percentage, then that lane has [a problem / issue]. R Turning; Rule 3: If X , Y There is a turning point S Turning, judging S Does the horizontal statistical proportion have the smallest percentage? If so, then... S If the steering is noise-driven, then determine... S Does the percentage of vertical statistics shift to greater than 1 / k , k This indicates the number of lanes in the guidance area; if so, then that lane exists. S Turning; (22) Middle lane identification rules Rule 4: If it exists L If the direction changes, then a judgment is made. L Does the percentage of vertical statistics shift to greater than 1 / k If so, then the lane exists. L If the steering wheel is turned, otherwise it will be rejected due to noise steering. If it exists R If the direction changes, then a judgment is made. R Does the percentage of vertical statistics shift to greater than 1 / k If so, then the lane exists. R If the steering wheel is turned, otherwise it will be rejected due to noise steering. Rule 5: If max( H X , H Y ) / ( H X + H Y (greater than) K 3. Then, the lane turn is identified as the turn with the largest lateral statistical proportion. If max( H X , H Y ) / ( H X + H Y (greater than) K 2, less than K 3. Then the lane has a turning point. X , Y Otherwise, the lane steering will be the least lateral steering; where H x H y These refer to the lateral statistics for X-turn and Y-turn, respectively. (23) Rightmost lane identification rules Rule 6: If X , Y There is a turning point R If you turn, then the lane exists. R Turning lane; Rule 7: If X , Y There is a turning point L If the lane is turning and has the largest longitudinal proportion, then that lane has... L Turning; Rule 8: If X , Y There is a turning point S Turning, judging S Does the horizontal statistical proportion have the smallest percentage? If so, then... S If the steering is noise-driven, then determine... S Does the percentage of vertical statistics shift to greater than 1 / k If so, then the lane exists. S Turn around.

2. The lane-level steering information detection method based on unsupervised learning as described in claim 1, characterized in that: Step 3.1 involves lateral statistics, specifically counting the number C of U-turn trajectories in the lane. D Number of left-turn trajectories C L Number of straight trajectories C S Number of right-turn trajectories C R The percentage of the total number of trajectories in this lane is specifically represented by H. D H L H S H R The specific calculation formula is as follows: (1) in ; For the first i Lane X Number of steering trajectories , where n is the number of lanes.

3. The lane-level steering information detection method based on unsupervised learning as described in claim 2, characterized in that: Step 3.1 Longitudinal statistics, i.e., counting the number C of U-turn trajectories in the lane. D The percentage of total U-turn trajectories and the number of left-turn trajectories in the guidance area (C) L The percentage of total left-turn trajectories and the number of straight-through trajectories in the guidance area (C) S The percentage of straight-ahead trajectories and the percentage of right-turn trajectories in the guidance area (C) R The percentage of the total number of right-turn trajectories in the guidance area is specifically represented by V. D V L V S V R The specific calculation formula is as follows: (2) in ; For the first i Lane X Number of steering trajectories , where n is the number of lanes.

4. The lane-level steering information detection method based on unsupervised learning as described in claim 1, characterized in that: The specific implementation method of step 3.2.2 is as follows: Rule 1: Count the total number of U-turn trajectories on the road segment. If it is less than... K 1. If U-turns are prohibited in all lanes of that road section, then if the value is greater than 1, then U-turns are prohibited in all lanes of that road section. K 1 indicates that a U-turn is permitted on this section of the road; Rule 2: If a U-turn is permitted on a road segment, and after noise-induced turn-by-turn elimination is performed on all lanes of that road segment, there are no remaining U-turn lanes, then the leftmost lane is assumed to be permitted for U-turns. Rule 3: If a lane allows U-turns, and after noise-driven turns are eliminated, the remaining turns in that lane are U-turn turns, and that turn has the largest proportion, then that lane can only allow U-turns; otherwise, that lane can allow U-turns.

5. A lane-level steering information detection system based on unsupervised learning, characterized in that, Includes the following modules: The preprocessing module is used for data preprocessing: first, it performs topology checks on the road network data and quality cleaning on the trajectory data; then, it performs map matching of the road network and trajectory data to connect the trajectory points with their respective road segments. The lane information recognition module is used for lane information recognition at intersections: First, the matching trajectory of the road segment within a few meters of the nearest endpoint of the intersection of the study road segment is denoised using a delanuary triangulation. Then, based on the Gaussian mixture model, cluster analysis is performed on the minimum distance from the matching trajectory point of the road segment to the road segment to obtain the number of lanes and lane line information of the intersection of the study road segment. The steering information detection module is used for lane-level steering information mining: first, it acquires lane steering trajectory information, then performs statistical analysis on the lane steering trajectory, and finally, based on different lane trajectory distributions, it detects lane-level steering information according to different rules. The specific implementation methods of the steering information detection module include: Step 3.1, Lane turning trajectory statistics: Count the number of U-turn (D), left turn (L), straight (S), and right turn (R) trajectories for each lane at the intersection, and then perform lateral and longitudinal statistics based on the number of each lane's turning trajectory. Step 3.2, Unsupervised lane turning identification: Noisy turns are eliminated, and then corresponding rules are designed based on the lateral and longitudinal statistical information of the remaining turns to detect lane turning information; The specific implementation method of step 3.2 is as follows: Step 3.2.1, noise removal: Turns with a horizontal statistical percentage of less than p% or a vertical statistical percentage of less than p% are considered noise turns and removed. Step 3.2.2, U-turn steering judgment: separate U-turns from other steering and perform lane-level steering information detection; Step 3.2.3, Other steering discrimination: Different rules are designed for the steering trajectory distribution of the leftmost, rightmost and middle lanes to identify lane steering information; The specific implementation method of step 3.2.3 is as follows: (1) If only one of the left turn, straight, and right turn statistics remains, then... X Then the lane must be turning. X Turning; (2) If there are no remaining statistical turns in the statistics, then the two turns X and Y with the largest values ​​in the horizontal statistics of left turn, straight, and right turn before noise removal are used to make the judgment; (3) If the number of remaining statistical turns is greater than or equal to 2, then the two turns X and Y with the largest remaining lateral statistics are used for discrimination; The specific implementation of the judgment based on the two largest lateral turns X and Y in steps (2) and (3) is as follows, where L is left turn, R is right turn, and S is straight turn: (21) Rule for determining the leftmost lane: Rule 1: If X , Y There is a turning point L If you turn, then the lane exists. L Turning lane; Rule 2: If X , Y There is a turning point R If the lane is turning and has the largest longitudinal percentage, then that lane has [a problem / issue]. R Turning; Rule 3: If X , Y There is a turning point S Turning, judging S Does the horizontal statistical proportion have the smallest percentage? If so, then... S If the steering is noise-driven, then determine... S Does the percentage of vertical statistics shift to greater than 1 / k , k This indicates the number of lanes in the guidance area; if so, then that lane exists. S Turning; (22) Middle lane identification rules Rule 4: If it exists L If the direction changes, then a judgment is made. L Does the percentage of vertical statistics shift to greater than 1 / k If so, then the lane exists. L If the steering wheel is turned, otherwise it will be rejected due to noise steering. If it exists R If the direction changes, then a judgment is made. R Does the percentage of vertical statistics shift to greater than 1 / k If so, then the lane exists. R If the steering wheel is turned, otherwise it will be rejected due to noise steering. Rule 5: If max( H X , H Y ) / ( H X + H Y (greater than) K 3. Then, the lane turn is identified as the turn with the largest lateral statistical proportion. If max( H X , H Y ) / ( H X + H Y (greater than) K 2, less than K 3. Then the lane has a turning point. X , Y Otherwise, the lane steering will be the least lateral steering; where H x H y These refer to the lateral statistics for X-turn and Y-turn, respectively. (23) Rightmost lane identification rules Rule 6: If X , Y There is a turning point R If you turn, then the lane exists. R Turning lane; Rule 7: If X , Y There is a turning point L If the lane is turning and has the largest longitudinal proportion, then that lane has... L Turning; Rule 8: If X , Y There is a turning point S Turning, judging S Does the horizontal statistical proportion have the smallest percentage? If so, then... S If the steering is noise-driven, then determine... S Does the percentage of vertical statistics shift to greater than 1 / k If so, then the lane exists. S Turn around.

Images