Road topological structure extraction method and system based on unmanned aerial vehicle data

Abstract

The invention discloses a road topological structure extraction method and system based on unmanned aerial vehicle data. The method comprises the steps of establishing transformation between a vehicle pixel coordinate system and a geographic coordinate system, inputting unmanned aerial vehicle image data, calculating geographic coordinates corresponding to vehicle pixel coordinates in each frame of image, and outputting the geographic coordinates as trajectory data; traversing the track data, executing a track smoothing algorithm, and calculating the motion direction and the motion state of the track by using the geometrical characteristics of the track; straight tracks are screened, the tracks are clustered based on the motion state and motion direction features of the vehicle, and representative tracks of each cluster in a clustering result are calculated; screening a non-straight track, and calculating an intersection driving area based on the non-straight track; and according to the representative track sequence and the intersection driving area, calculating to obtain a complete topological relation of the intersection lane. The method can realize quasi real-time updating of the road topological structure, and is suitable for quick response of dynamic scenes such as construction areas and temporary traffic organization adjustment.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and intelligent transportation, specifically relating to a method and system for extracting road topology based on UAV aerial image data. Background Technology

[0002] Existing high-precision electronic maps can provide relatively complete topology structures at the lane level, but their construction relies on dedicated data collection vehicles and high-precision surveying equipment, resulting in high maintenance costs. Furthermore, high-precision maps have long update cycles, making it difficult to reflect real-time changes at intersections caused by construction, temporary traffic organization adjustments, etc. While drones are lightweight and highly mobile, capable of acquiring intersection images in real time, their imagery data cannot be directly used for lane-level topology calculations.

[0003] With the widespread adoption of vehicle positioning and sensing devices, the industry is attempting to reconstruct intersection topology using vehicle trajectory data. This method can reflect the actual path relationships of vehicles to a certain extent, and is closer to real-world usage scenarios than static maps. However, because trajectory collection relies on the large-scale accumulation of data from vehicle terminals, and trajectory points may suffer from insufficient sampling accuracy, noise interference, and uneven distribution, the practical application effect is often limited, especially in areas lacking a large number of sampled vehicles, making it difficult to obtain a complete intersection topology description in a timely manner.

[0004] Drones offer advantages such as flexible deployment, wide coverage, and high data collection efficiency; by collecting overhead images of intersections, they can provide a comprehensive view of traffic conditions. However, currently, drone imagery data is rarely directly applied to lane-level topology recognition, and is more often used for 3D modeling or road segment mapping. Meanwhile, traditional distance measurement methods for trajectories, such as Hausdorff distance or Fraser distance, cannot accurately extract the trajectory characteristics of vehicles within the intersection area. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, the present invention aims to provide a method and system for extracting road topology based on UAV data. This method utilizes UAV video streams for target detection and performs inverse projection transformation on UAV images to calculate the trajectories of vehicles at intersections within a certain time window. The present invention can solve the problems of high cost and delayed updates in high-precision map construction, as well as data limitations in vehicle trajectory-based methods in the existing technologies.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] The present invention provides a method for extracting road topology based on UAV data, comprising the following steps:

[0008] 1) Establish the transformation between the vehicle pixel coordinate system and the geographic coordinate system. Input UAV image data, calculate the geographic coordinates corresponding to the vehicle pixel coordinates in each frame of the image, and output the trajectory data.

[0009] 2) Traverse trajectory data The trajectory smoothing algorithm is executed to calculate the trajectory's direction and state of motion using the trajectory's geometric features;

[0010] 3) Filter straight-line trajectories, cluster the trajectories based on the vehicle's motion state and direction characteristics, and calculate the representative trajectory of each cluster in the clustering results;

[0011] 4) Filter out non-straight-ahead trajectories and calculate the intersection driving area based on the non-straight-ahead trajectories. ;

[0012] 5) Based on the representative trajectory sequence and intersection driving area The complete topological relationship of the lanes at the intersection is calculated.

[0013] Further, step 1) specifically includes:

[0014] 11) If M trajectories are detected within one flight cycle of the UAV, let the input sequence be denoted as . ,in, This represents the sequence number of the i-th vehicle. Let the trajectory sequence corresponding to the i-th vehicle be defined as follows: , Indicates the first The length of each trajectory. Coordinates in the vehicle pixel coordinate system; camera intrinsics of the UAV. , and For camera focal length parameters, and The coordinate parameters of the principal point in the image; the extrinsic parameters of the UAV are... R represents the UAV rotation matrix. This indicates the position of the camera origin in the world coordinate system; the corresponding coordinates in the world coordinate system are: Calculate world coordinate system coordinates as follows:

[0015] ;

[0016] in, This represents the distance from a 3D point to the center of the camera in the camera coordinate system.

[0017] 12) The obtained data will be obtained through georeferencing. Convert the WGS84 latitude and longitude data to a geographic coordinate system and output a set of trajectory data. ,in Indicates the first Group trajectory data.

[0018] Further, step 2) specifically includes:

[0019] 21) Traverse trajectory data Calculate the UTM coordinates of the trajectory ;

[0020] 22) Traverse trajectory data , for trajectory Perform velocity filtering; calculate the instantaneous velocity between adjacent trajectory points. The expression is as follows:

[0021] ;

[0022] in, This represents the time interval for trajectory sampling, which is set to 25ms by default. , () represents the planar coordinates of the i-th trajectory point in the UTM coordinate system. , () represents the planar coordinates of the (i-1)th trajectory point in the UTM coordinate system;

[0023] Set speed threshold Eliminate those that meet the requirements Low-speed anomalies were identified, and the filtered trajectory was obtained. ; for trajectory A two-dimensional Kalman filter is applied to smooth the trajectory, resulting in a smoothed trajectory. and output the smoothed trajectory set. ;

[0024] 23) Calculate the motion state Traversal trajectory Take the average unit vector of the first third of the trajectory The average unit vector of the latter third of the trajectory Calculate the angle between two vectors. , The trajectory motion states are divided into six types based on the angle range: straight, left turn, right turn, U-shape, stationary, and unknown.

[0025] 24) Calculate the direction of motion After taking Calculate the average unit direction vector of the second half of the trajectory at each point. The vehicle's heading is determined based on the quadrant and dominance of the directional components; the geographic azimuth is calculated. as follows:

[0026] ;

[0027] in, The unit component representing the trajectory in the eastward direction; The unit component representing the trajectory in the north direction;

[0028] Based on the geographical azimuth range, the directions of movement are divided into east, south, west, north, northeast, southeast, northwest, and southwest;

[0029] 25) Combine the motion state and direction to output a set of state data representing the vehicle's trajectory. ,in These represent the vehicle's serial number, motion status label, driving direction label, and original trajectory data, respectively.

[0030] Furthermore, in step 23), the average unit vector of the trajectory is calculated using the cumulative average direction vector method, specifically: the trajectory point sequence is divided into three segments, and the first segment is taken... The first segment consists of several points, followed by... Each point forms a termination segment. The displacement vectors of all adjacent points within each segment are accumulated and normalized to obtain the unit direction vector. If the cumulative displacement is too small, the corresponding direction vector is set to zero.

[0031] Furthermore, step 3) specifically includes:

[0032] 31) Filter straight-line trajectories from the state data, selecting a subset of trajectories whose motion state is straight-line and whose direction of travel is consistent. ;

[0033] 32) Traverse each set of straight-line trajectories subset Calculate the distances between the internal trajectories to obtain the distance matrix. ;

[0034] 33) Employ a hierarchical clustering algorithm to iteratively merge the nearest clusters and select the distance matrix. As input, the final clustering result is output. , K is the number of clusters, and each cluster represents a lane;

[0035] 34) For each cluster Select its centroid trajectory As a representative trajectory, this representative trajectory is the trajectory with the smallest average distance to all other trajectories in the cluster;

[0036] 35) For the representative trajectory set Perform geometric calibration; establish a geometric reference representing the trajectory. Geometric centroid The starting and ending points of the trajectory are projected onto the reference line, and the range of projection parameters is calculated. Let represent the minimum and maximum values ​​of the projection parameters obtained after projecting the start and end points of each trajectory onto the geometric reference line representing the trajectory, respectively; based on the geometric reference and the projection range, calculate the start point of the representative trajectory. and the end point as follows:

[0037] ;

[0038] ;

[0039] The final representative trajectory is denoted as Output represents the trajectory sequence .

[0040] Furthermore, in step 32), the trajectory subset... Calculate any two different trajectories in sequence. and Total distance Specifically, it includes:

[0041] 321) Calculate the angular distance : on trajectory and Take its first point and last point respectively to form the head vector Sum Tail Vector Angular distance is calculated using the inverse cosine function. as follows:

[0042] ;

[0043] 322) Calculate the vertical distance Calculate separately The first point, tail point, The first point, The Euclidean distance from the tail point to the straight line of the opponent's trajectory is denoted as . The weighted average yields the vertical distance. as follows:

[0044] ;

[0045] 323) Calculate the overlap distance :pass Trajectory directly projected to Trajectory calculation overlap length The overlap distance is The calculation formula is:

[0046] ;

[0047] in, , They are respectively and The length of the trajectory;

[0048] 324) Weighted distance Traverse all trajectories and output the distance matrix. ,in , and Angular distance Vertical distance Overlap distance The weighting coefficients, .

[0049] Furthermore, the intersection driving area in step 4) is a geometric area that completely covers the range of straight-going and turning movements of vehicles within the intersection.

[0050] Further, step 4) specifically includes:

[0051] 41) From state data The set of trajectories whose motion state is not straight is selected from the middle. ;

[0052] 42) Traversing the set of trajectories For each trajectory Calculate the angle θ between two adjacent vectors sequentially, and take the point with the largest angle θ value in the trajectory as the maximum turning point of the trajectory. The set of maximum turning points is denoted as , , representing the point with the largest turning angle for the vehicle under each trajectory;

[0053] 43) Calculate the set of maximum turning points geometric center and the set of maximum turning points From each point in the middle to the center The maximum distance is used as the radius The expression is as follows:

[0054] ;

[0055] in, The function for calculating Euclidean distance;

[0056] Build with Center of the circle A circular area with a radius of 1 is designated as the driving area at the intersection. .

[0057] Furthermore, step 5) specifically includes:

[0058] 51) Calculate the intersection of the driving area at the intersection and the representative trajectory, and then truncate the representative trajectory into the entering portion. Intersection With the exit section ;Pick and Represents the lanes at this intersection;

[0059] 52) Traversal and Perform buffer analysis on the target trajectory and output the lane buffer. ,in This represents the buffer zone for the i-th lane; the expression is as follows:

[0060] ;

[0061] in, Lane width, For the first Center line of lane, buffer function Used to equidistantly expand line features in geometric space to generate polygonal regions representing the actual driving space of a lane;

[0062] 53) Utilize lane buffer zones and trajectory data The final topological relationships are calculated; the number of topological relationships is set to be... The final output set of topological relationships. for:

[0063] ;

[0064] in, For the first The entrance lane identifier corresponding to each topological relationship. For the first The exit lane sign corresponding to each topological relationship. Let be the set of identifiers for the entrance lanes in any topological relationship. Let be the set of identifiers for the exit lanes in any topological relationship;

[0065] For each trajectory The first half of the trajectory and the second half are respectively and The topology calculation formula is:

[0066] ;

[0067] ;

[0068] Topological relation set Characterizes the road topology of intersections.

[0069] The present invention also provides a road topology extraction system based on UAV data, comprising:

[0070] The output module is used to establish the transformation between the vehicle pixel coordinate system and the geographic coordinate system. It takes UAV image data as input, calculates the geographic coordinates corresponding to the vehicle pixel coordinates in each frame of the image, and outputs trajectory data.

[0071] The first calculation module is used to traverse the trajectory data. The trajectory smoothing algorithm is executed to calculate the trajectory's direction and state of motion using the trajectory's geometric features;

[0072] The second calculation module is used to filter straight-line trajectories, cluster the trajectories based on the vehicle's motion state and direction features, and calculate the representative trajectory of each cluster in the clustering results.

[0073] The third calculation module is used to filter non-straight-ahead trajectories and calculate the intersection driving area based on the non-straight-ahead trajectories. ;

[0074] The fourth calculation module is used to calculate based on the representative trajectory sequence. and intersection driving area The complete topological relationship of the lanes at the intersection is calculated.

[0075] The beneficial effects of this invention are:

[0076] This invention avoids the need for dedicated data collection vehicles and high-precision surveying equipment, thereby reducing the cost of acquiring and updating road topology data.

[0077] This invention enables near real-time updates of road topology, making it suitable for rapid response in dynamic scenarios such as construction zones and temporary traffic organization adjustments.

[0078] Based on the characteristics of intersection trajectory data, this invention designs an improved trajectory distance metric algorithm for trajectory clustering, which improves the clustering effect and enables more accurate clustering of relevant representative trajectories. Attached Figure Description

[0079] Figure 1 This is a schematic diagram of the method of the present invention.

[0080] Figure 2 A schematic diagram of converting UAV imagery data to the world coordinate system.

[0081] Figure 3 This is a schematic diagram of the output trajectory data.

[0082] Figure 4 This is a schematic diagram for calculating lane topology. Detailed Implementation

[0083] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0084] Reference Figure 1 As shown, a method for extracting road topology based on UAV data includes the following steps:

[0085] 1) Establish the transformation between the vehicle pixel coordinate system and the geographic coordinate system. Input UAV imagery data, calculate the geographic coordinates corresponding to the vehicle pixel coordinates in each frame of the image, and output trajectory data; specifically including:

[0086] 11) Reference Figure 2 As shown, M trajectories are detected within one flight cycle of the UAV. Let the input sequence be... ,in, This represents the sequence number of the i-th vehicle. Let the trajectory sequence corresponding to the i-th vehicle be defined as follows: , Indicates the first The length of each trajectory. Coordinates in the vehicle pixel coordinate system; camera intrinsics of the UAV. , and For camera focal length parameters, and The coordinate parameters of the principal point in the image; the extrinsic parameters of the UAV are... R represents the UAV rotation matrix. This indicates the position of the camera origin in the world coordinate system; the corresponding coordinates in the world coordinate system are: Calculate world coordinate system coordinates as follows:

[0087] ;

[0088] in, This represents the distance from a 3D point to the center of the camera in the camera coordinate system.

[0089] 12) Reference Figure 3 As shown, the obtained results will be obtained through georeferencing. Convert the WGS84 latitude and longitude data to a geographic coordinate system and output a set of trajectory data. ,in Indicates the first Group trajectory data.

[0090] 2) Traverse trajectory data The trajectory smoothing algorithm is executed, which calculates the trajectory's direction and state of motion using the trajectory's geometric features; specifically, it includes:

[0091] 21) Traverse trajectory data Calculate the UTM coordinates of the trajectory ;

[0092] 22) Traverse trajectory data , for trajectory Perform velocity filtering; calculate the instantaneous velocity between adjacent trajectory points. The expression is as follows:

[0093] ;

[0094] in, This represents the time interval for trajectory sampling, which is set to 25ms by default. , () represents the planar coordinates of the i-th trajectory point in the UTM coordinate system. , () represents the planar coordinates of the (i-1)th trajectory point in the UTM coordinate system;

[0095] Set speed threshold Eliminate those that meet the requirements Low-speed anomalies were identified, and the filtered trajectory was obtained. ; for trajectory A two-dimensional Kalman filter is applied to smooth the trajectory, resulting in a smoothed trajectory. and output the smoothed trajectory set. ;

[0096] 23) Calculate the motion state Traversal trajectory Take the average unit vector of the first third of the trajectory The average unit vector of the latter third of the trajectory Calculate the angle between two vectors. , The trajectory motion states are divided into six types based on the angle range: straight, left turn, right turn, U-shape, stationary, and unknown.

[0097] 24) Calculate the direction of motion After taking Calculate the average unit direction vector of the second half of the trajectory at each point. The vehicle's heading is determined based on the quadrant and dominance of the directional components; the geographic azimuth is calculated. as follows:

[0098] ;

[0099] in, The unit component representing the trajectory in the eastward direction; The unit component representing the trajectory in the north direction;

[0100] Based on the geographical azimuth range, the directions of movement are divided into east, south, west, north, northeast, southeast, northwest, and southwest;

[0101] 25) Combine the motion state and direction to output a set of state data representing the vehicle's trajectory. ,in These represent the vehicle's serial number, motion status label, driving direction label, and original trajectory data, respectively.

[0102] Specifically, in step 23), the average unit vector of the trajectory is calculated using the cumulative average direction vector method, which involves dividing the trajectory point sequence into three segments and taking the first segment... The first segment consists of several points, followed by... Each point forms a termination segment. The displacement vectors of all adjacent points within each segment are accumulated and normalized to obtain the unit direction vector. If the cumulative displacement is too small, the corresponding direction vector is set to zero.

[0103] 3) Filter straight-line trajectories, cluster the trajectories based on the vehicle's motion state and direction characteristics, and calculate the representative trajectory of each cluster in the clustering results; specifically including:

[0104] 31) Filter straight-line trajectories from the state data, selecting a subset of trajectories whose motion state is straight-line and whose direction of travel is consistent. ;

[0105] 32) Traverse each set of straight-line trajectories subset Calculate the distances between the internal trajectories to obtain the distance matrix. ;

[0106] 33) Employ a hierarchical clustering algorithm to iteratively merge the nearest clusters and select the distance matrix. As input, the final clustering result is output. , K is the number of clusters, and each cluster represents a lane;

[0107] 34) For each cluster Select its centroid trajectory As a representative trajectory, this representative trajectory is the trajectory with the smallest average distance to all other trajectories in the cluster;

[0108] 35) For the representative trajectory set Perform geometric calibration; establish a geometric reference representing the trajectory. Geometric centroid The starting and ending points of the trajectory are projected onto the reference line, and the range of projection parameters is calculated. Let represent the minimum and maximum values ​​of the projection parameters obtained after projecting the start and end points of each trajectory onto the geometric reference line representing the trajectory, respectively; based on the geometric reference and the projection range, calculate the start point of the representative trajectory. and the end point as follows:

[0109] ;

[0110] ;

[0111] The final representative trajectory is denoted as Output represents the trajectory sequence .

[0112] Specifically, in step 32), the trajectory subset... Calculate any two different trajectories in sequence. and Total distance Specifically, it includes:

[0113] 321) Calculate the angular distance : on trajectory and Take its first point and last point respectively to form the head vector Sum Tail Vector Angular distance is calculated using the inverse cosine function. as follows:

[0114] ;

[0115] 322) Calculate the vertical distance Calculate separately The first point, tail point, The first point, The Euclidean distance from the tail point to the straight line of the opponent's trajectory is denoted as . The weighted average yields the vertical distance. as follows:

[0116] ;

[0117] 323) Calculate the overlap distance :pass Trajectory directly projected to Trajectory calculation overlap length The overlap distance is The calculation formula is:

[0118] ;

[0119] in, , They are respectively and The length of the trajectory;

[0120] 324) Weighted distance Traverse all trajectories and output the distance matrix. ,in , and Angular distance Vertical distance Overlap distance The weighting coefficients, .

[0121] 4) Filter out non-straight-ahead trajectories and calculate the intersection driving area based on the non-straight-ahead trajectories. ;

[0122] In step 4), the intersection driving area is a geometric area that completely covers the range of straight-going and turning movements of vehicles within the intersection.

[0123] Step 4) specifically includes:

[0124] 41) From state data The set of trajectories whose motion state is not straight is selected from the middle. ;

[0125] 42) Traversing the set of trajectories For each trajectory Calculate the angle θ between two adjacent vectors sequentially, and take the point with the largest angle θ value in the trajectory as the maximum turning point of the trajectory. The set of maximum turning points is denoted as , , representing the point with the largest turning angle for the vehicle under each trajectory;

[0126] 43) Calculate the set of maximum turning points geometric center and the set of maximum turning points From each point in the middle to the center The maximum distance is used as the radius The expression is as follows:

[0127] ;

[0128] in, The function for calculating Euclidean distance;

[0129] Build with Center of the circle A circular area with a radius of 1 is designated as the driving area at the intersection. .

[0130] 5) Based on the representative trajectory sequence and intersection driving area The complete topological relationship of the intersection lanes is calculated; specifically including:

[0131] 51) Calculate the intersection of the driving area at the intersection and the representative trajectory, and then truncate the representative trajectory into the entering portion. Intersection With the exit section ;Pick and Represents the lanes at this intersection;

[0132] 52) Traversal and Perform buffer analysis on the target trajectory and output the lane buffer. ,in This represents the buffer zone for the i-th lane; the expression is as follows:

[0133] ;

[0134] in, Lane width, For the first Center line of lane, buffer function Used to equidistantly expand line features in geometric space to generate polygonal regions representing the actual driving space of a lane;

[0135] 53) Reference Figure 4 As shown, lane buffer zones are utilized. and trajectory data The final topological relationships are calculated; the number of topological relationships is set to be... The final output set of topological relationships. for:

[0136] ;

[0137] in, For the first The entrance lane identifier corresponding to each topological relationship. For the first The exit lane sign corresponding to each topological relationship. Let be the set of identifiers for the entrance lanes in any topological relationship. Let be the set of identifiers for the exit lanes in any topological relationship;

[0138] For each trajectory The first half of the trajectory and the second half are respectively and The topology calculation formula is:

[0139] ;

[0140] ;

[0141] Topological relation set Characterizes the road topology of intersections.

[0142] The present invention also provides a road topology extraction system based on UAV data, comprising:

[0143] The output module is used to establish the transformation between the vehicle pixel coordinate system and the geographic coordinate system. It takes UAV image data as input, calculates the geographic coordinates corresponding to the vehicle pixel coordinates in each frame of the image, and outputs trajectory data.

[0144] The first calculation module is used to traverse the trajectory data. The trajectory smoothing algorithm is executed to calculate the trajectory's direction and state of motion using the trajectory's geometric features;

[0145] The second calculation module is used to filter straight-line trajectories, cluster the trajectories based on the vehicle's motion state and direction features, and calculate the representative trajectory of each cluster in the clustering results.

[0146] The third calculation module is used to filter non-straight-ahead trajectories and calculate the intersection driving area based on the non-straight-ahead trajectories. ;

[0147] The fourth calculation module is used to calculate based on the representative trajectory sequence. and intersection driving area The complete topological relationship of the lanes at the intersection is calculated.

[0148] This invention has many specific applications. The above description is only a preferred embodiment of this invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of this invention, and these improvements should also be considered within the scope of protection of this invention.

Claims

1. A method for extracting road topology based on UAV data, characterized in that, The steps are as follows: 1) Establish the transformation between the vehicle pixel coordinate system and the geographic coordinate system. Input UAV image data, calculate the geographic coordinates corresponding to the vehicle pixel coordinates in each frame of the image, and output the trajectory data. 2) Traverse trajectory data The trajectory smoothing algorithm is executed to calculate the trajectory's direction and state of motion using the trajectory's geometric features; 3) Filter straight-line trajectories, cluster the trajectories based on the vehicle's motion state and direction characteristics, and calculate the representative trajectory of each cluster in the clustering results; 4) Filter out non-straight-ahead trajectories and calculate the intersection driving area based on the non-straight-ahead trajectories. ; 5) Based on the representative trajectory sequence and intersection driving area The complete topological relationship of the lanes at the intersection is calculated.

2. The road topology extraction method based on UAV data according to claim 1, characterized in that, Step 1) specifically includes: 11) If M trajectories are detected within one flight cycle of the UAV, let the input sequence be denoted as . ,in, This represents the sequence number of the i-th vehicle. Let the trajectory sequence corresponding to the i-th vehicle be defined as follows: , Indicates the first The length of each trajectory. Coordinates in the vehicle pixel coordinate system; camera intrinsics of the UAV. , and For camera focal length parameters, and The coordinate parameters of the principal point in the image; the extrinsic parameters of the UAV are... R represents the UAV rotation matrix. This indicates the position of the camera origin in the world coordinate system; the corresponding coordinates in the world coordinate system are: Calculate world coordinate system coordinates as follows: ； in, This represents the distance from a 3D point to the center of the camera in the camera coordinate system. 12) The obtained data will be obtained through georeferencing. Convert the WGS84 latitude and longitude data to a geographic coordinate system and output a set of trajectory data. ,in Indicates the first Group trajectory data.

3. The road topology extraction method based on UAV data according to claim 1, characterized in that, Step 2) specifically includes: 21) Traverse trajectory data Calculate the UTM coordinates of the trajectory ; 22) Traverse trajectory data , for trajectory Perform velocity filtering; calculate the instantaneous velocity between adjacent trajectory points. The expression is as follows: ； in, Indicates the time interval for trajectory sampling; , () represents the planar coordinates of the i-th trajectory point in the UTM coordinate system. , () represents the planar coordinates of the (i-1)th trajectory point in the UTM coordinate system; Set speed threshold Eliminate those that meet the requirements Low-speed anomalies were identified, and the filtered trajectory was obtained. ; for trajectory A two-dimensional Kalman filter is applied to smooth the trajectory, resulting in a smoothed trajectory. and output the smoothed trajectory set. ; 23) Calculate the motion state Traversal trajectory Take the average unit vector of the first third of the trajectory The average unit vector of the latter third of the trajectory Calculate the angle between two vectors. , The trajectory motion states are divided into six types based on the angle range: straight, left turn, right turn, U-shape, stationary, and unknown. 24) Calculate the direction of motion After taking Calculate the average unit direction vector of the second half of the trajectory at each point. The vehicle's heading is determined based on the quadrant and dominance of the directional components; the geographic azimuth is calculated. as follows: ； in, The unit component representing the trajectory in the eastward direction; The unit component representing the trajectory in the north direction; Based on the geographical azimuth range, the directions of movement are divided into east, south, west, north, northeast, southeast, northwest, and southwest; 25) Combine the motion state and direction to output a set of state data representing the vehicle's trajectory. ,in These represent the vehicle's serial number, motion status label, driving direction label, and original trajectory data, respectively.

4. The road topology extraction method based on UAV data according to claim 3, characterized in that, In step 23), the average unit vector of the trajectory is calculated using the cumulative average direction vector method, specifically: the trajectory point sequence is divided into three segments, and the first segment is taken... The first segment consists of several points, followed by... Each point forms a termination segment. The displacement vectors of all adjacent points within each segment are accumulated and normalized to obtain the unit direction vector. If the cumulative displacement is too small, the corresponding direction vector is set to zero.

5. The method for extracting road topology based on UAV data according to claim 1, characterized in that, Step 3) specifically includes: 31) Filter straight-line trajectories from the state data, selecting a subset of trajectories whose motion state is straight-line and whose direction of travel is consistent. ; 32) Traverse each set of straight-line trajectories subset Calculate the distances between the internal trajectories to obtain the distance matrix. ; 33) Employ a hierarchical clustering algorithm to iteratively merge the nearest clusters and select the distance matrix. As input, the final clustering result is output. , K is the number of clusters, and each cluster represents a lane; 34) For each cluster Select its centroid trajectory As a representative trajectory, this representative trajectory is the trajectory with the smallest average distance to all other trajectories in the cluster; 35) For the representative trajectory set Perform geometric calibration; establish a geometric reference representing the trajectory. Geometric centroid The starting and ending points of the trajectory are projected onto the reference line, and the range of projection parameters is calculated. Let represent the minimum and maximum values ​​of the projection parameters obtained after projecting the start and end points of each trajectory onto the geometric reference line representing the trajectory, respectively; based on the geometric reference and the projection range, calculate the start point of the representative trajectory. and the end point as follows: ； ； The final representative trajectory is denoted as Output represents the trajectory sequence .

6. The method for extracting road topology based on UAV data according to claim 5, characterized in that, In step 32), the trajectory subset Calculate any two different trajectories in sequence. and Total distance Specifically, it includes: 321) Calculate the angular distance : on trajectory and Take its first point and last point respectively to form the head vector Sum Tail Vector Angular distance is calculated using the inverse cosine function. as follows: ； 322) Calculate the vertical distance Calculate separately The first point, tail point, The first point, The Euclidean distance from the tail point to the straight line of the opponent's trajectory is denoted as . The weighted average yields the vertical distance. as follows: ； 323) Calculate the overlap distance :pass Trajectory directly projected to Trajectory calculation overlap length The overlap distance is The calculation formula is: ； in, , They are respectively and The length of the trajectory; 324) Weighted distance Traverse all trajectories and output the distance matrix. ,in , and Angular distance Vertical distance Overlap distance The weighting coefficients, .

7. The method for extracting road topology based on UAV data according to claim 1, characterized in that, The intersection driving area in step 4) is a geometric area that completely covers the range of straight-going and turning movements of vehicles within the intersection.

8. The method for extracting road topology based on UAV data according to claim 1, characterized in that, Step 4) specifically includes: 41) From state data The set of trajectories whose motion state is not straight is selected from the middle. ; 42) Traversing the set of trajectories For each trajectory Calculate the angle θ between two adjacent vectors sequentially, and take the point with the largest angle θ value in the trajectory as the maximum turning point of the trajectory. The set of maximum turning points is denoted as , , representing the point with the largest turning angle for the vehicle under each trajectory; 43) Calculate the set of maximum turning points geometric center and the set of maximum turning points From each point in the middle to the center The maximum distance is used as the radius The expression is as follows: ； in, The function for calculating Euclidean distance; Build with Center of the circle A circular area with a radius of 1 is designated as the driving area at the intersection. .

9. The method for extracting road topology based on UAV data according to claim 1, characterized in that, Step 5) specifically includes: 51) Calculate the intersection of the driving area at the intersection and the representative trajectory, and then truncate the representative trajectory into the entering portion. Intersection With the exit section ;Pick and Represents the lanes at this intersection; 52) Traversal and Perform buffer analysis on the target trajectory and output the lane buffer. ,in This represents the buffer zone for the i-th lane; the expression is as follows: ； in, Lane width, For the first Center line of lane, buffer function Used to equidistantly expand line features in geometric space to generate polygonal regions representing the actual driving space of a lane; 53) Utilize lane buffer zones and trajectory data The final topological relationships are calculated; the number of topological relationships is set to be... The final output set of topological relationships. for: ； in, For the first The entrance lane identifier corresponding to each topological relationship. For the first The exit lane sign corresponding to each topological relationship. Let be the set of identifiers for the entrance lanes in any topological relationship. Let be the set of identifiers for the exit lanes in any topological relationship; For each trajectory The first half of the trajectory and the second half are respectively and The topology calculation formula is: ； ； Topological relation set Characterizes the road topology of intersections.

10. A road topology extraction system based on UAV data, characterized in that, include: The output module is used to establish the transformation between the vehicle pixel coordinate system and the geographic coordinate system. It takes UAV image data as input, calculates the geographic coordinates corresponding to the vehicle pixel coordinates in each frame of the image, and outputs trajectory data. The first calculation module is used to traverse the trajectory data. The trajectory smoothing algorithm is executed to calculate the trajectory's direction and state of motion using the trajectory's geometric features; The second calculation module is used to filter straight-line trajectories, cluster the trajectories based on the vehicle's motion state and direction features, and calculate the representative trajectory of each cluster in the clustering results. The third calculation module is used to filter non-straight-ahead trajectories and calculate the intersection driving area based on the non-straight-ahead trajectories. ; The fourth calculation module is used to calculate based on the representative trajectory sequence. and intersection driving area The complete topological relationship of the lanes at the intersection is calculated.

Images