Mapping method and system

By obtaining the trace map and laser point cloud data of the unmanned vehicle, and using the inflection point extraction and fitting algorithm to generate Opendrive high-precision maps, the problem of lack of environmental information in the unmanned vehicle map is solved, and low-cost, simple and efficient construction is achieved The diagram process quickly supplements traffic lights and traffic sign information.

CN118980369APending Publication Date: 2024-11-19BEIJING EGOVA TECH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411072439.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

Existing unmanned vehicles lack detailed environmental information such as road characteristics and traffic signs when relying on trace-seeking maps. Moreover, high-precision map production costs are high. Relying on professional equipment and manpower, it is difficult to achieve low-cost solutions for update and maintenance.

Method used

By obtaining trace-seeking map data and laser point cloud data, extracting inflection points, fitting straight lines and curves, using cubic Bezier curves and particle swarm optimization algorithms, generating road reference lines and lane data, quickly supplement traffic lights and traffic sign information, and outputting Opendrive high-precision map.

Benefits of technology

It realizes low-cost, simple and efficient map construction of unmanned vehicle maps, reduces the requirements for lidar equipment, quickly supplements the missing environmental information in the trace search map, and provides a comprehensive reference for map element collection and inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118980369A_ABST
    Figure CN118980369A_ABST
Patent Text Reader

Abstract

The invention discloses a mapping method and system, and the method comprises the following steps: obtaining tracing map data, and generating an original track; acquiring laser point cloud data, and constructing a point cloud model for display; obtaining a road reference line according to the original track and the laser point cloud data; after the road reference line is obtained, lane data is obtained by using the laser point cloud data; after lane data are generated, intersection acquisition is carried out; and after intersection acquisition is completed, signal lamps and traffic marking lines are acquired according to the laser point cloud data, and a map is completed and exported. According to the method, the Opendrive map can be established only through the track point data and the laser point cloud data, the requirement for laser radar equipment is not high, 16-line radar or 32-line radar can be achieved, the mapping process is simple and efficient, and the cost is low.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular, to a mapping method and system thereof. Background Art

[0002] For small-scale unmanned vehicles used for logistics, patrols, etc. or in closed parks, they are relatively small in size, slow in speed, and have relatively fixed routes. The autonomous driving solutions for such unmanned vehicles are mainly based on tracking maps and high-precision maps. Tracking maps provide simplified path guidance for unmanned vehicles, focusing on defining the vehicle's driving path. The accuracy requirement is not high. It only needs to use GPS and IMU and other sensors to assist in improving positioning accuracy to record the actual driving path of the unmanned vehicle to complete the production of tracking maps. The production cost is low. High-precision maps provide unmanned vehicles with detailed traffic information, such as lane lines, traffic signs, curbs, intersection structures, road slopes, etc., as well as dynamically updated traffic rules information. The production of high-precision maps requires professional collection vehicles equipped with laser radar, GPS, RTK, IMU, high-definition cameras and other equipment. It may also rely on multi-source data fusion such as satellite images, street view images, drone aerial photography, and then perform a series of internal processing processes such as point cloud stitching, base map production, feature extraction, and feature annotation. However, the tracking map can only provide macro-path guidance, lacking detailed environmental information such as road features and traffic signs, which limits the unmanned vehicle's in-depth understanding of the surrounding environment and decision-making ability. It cannot flexibly respond to sudden obstacle information and can only rely entirely on the sensor's perception ability to avoid obstacles. In addition, in environments with weak or no GPS signals, such as tunnels and high-rise dense areas, the performance of the tracking map may drop significantly. The production of high-precision maps requires professional data acquisition equipment, professional data processing software and manual processing. The entire process is highly dependent on automated processing tools and artificial intelligence algorithms, and the production cost is very high. In addition, the subsequent map update and maintenance also requires a lot of manpower and material resources.

[0003] Therefore, how to provide a solution to the defect of unmanned vehicles relying on tracking maps, which lacks detailed environmental information such as road features and traffic signs, while achieving a low-cost mapping method, has become an urgent problem to be solved in this field. Summary of the invention

[0004] The present application proposes a mapping method, comprising the following steps: obtaining tracking map data and generating an original trajectory; obtaining laser point cloud data and constructing a point cloud model for display; obtaining a road reference line based on the original trajectory and the laser point cloud data; after completing the acquisition of the road reference line, obtaining lane data using the laser point cloud data; after generating the lane data, performing intersection collection; after completing the intersection collection, collecting signal lights and traffic markings based on the laser point cloud data, and completing and exporting the map.

[0005] As mentioned above, obtaining the tracking map data includes importing the unmanned vehicle tracking map, drawing the track points in sequence and generating the original track.

[0006] As above, according to the original trajectory and laser point cloud data, obtaining the road reference line includes the following sub-steps: extracting inflection points; fitting straight line segments after inflection point extraction; fitting curve segments after straight line fitting; connecting each segment of the fitting trajectory in turn to obtain a fitted trajectory line; translating the fitted trajectory line by a specified distance to obtain a road reference line.

[0007] As above, the inflection point extraction includes determining the curvature of each trajectory point, where the curvature c is expressed as: The first-order derivative value y'=2ax+b, the second-order derivative value y"=2a, a and b are derivative coefficients; when the curvature c value is greater than the specified threshold, the current trajectory point is an inflection point.

[0008] As above, the straight line segment fitting includes performing least squares straight line fitting on every two inflection points on the straight road and the trajectory points between the inflection points, obtaining continuous straight line segments after fitting, calculating the angle between the direction vectors of adjacent straight line segments, and merging the adjacent straight line segments into one line segment if the angle does not exceed a specified threshold.

[0009] A mapping system specifically includes: an original trajectory generation unit, a model construction unit, a road reference line acquisition unit, a lane data acquisition unit, an intersection acquisition unit, and a signal light and traffic line acquisition unit; the original trajectory generation unit is used to acquire tracking map data and generate an original trajectory; the model construction unit is used to acquire laser point cloud data and build a point cloud model for display; the road reference line acquisition unit is used to acquire road reference lines according to the original trajectory and laser point cloud data; the lane data acquisition unit is used to obtain lane data using laser point cloud data; the intersection acquisition unit is used to perform intersection acquisition; the signal light and traffic line acquisition unit is used to collect signal lights and traffic lines according to the laser point cloud data, and complete and export the map.

[0010] As mentioned above, the original trajectory generation unit obtains the tracking map data, including importing the unmanned vehicle tracking map, drawing the trajectory points in sequence and then generating the original trajectory.

[0011] As above, the road reference line acquisition unit acquires the road reference line according to the original trajectory and the laser point cloud data, and includes the following sub-steps: extracting inflection points; fitting straight line segments after inflection point extraction; fitting curve segments after straight line fitting; connecting each segment of the fitting trajectory in sequence to obtain a fitted trajectory line; translating the fitted trajectory line by a specified distance to obtain a road reference line.

[0012] As above, the inflection point extraction in the road reference line acquisition unit includes determining the curvature of each trajectory point, wherein the curvature c is expressed as: The first-order derivative value y'=2ax+b, the second-order derivative value y"=2a, a and b are derivative coefficients; when the curvature c value is greater than the specified threshold, the current trajectory point is an inflection point.

[0013] As mentioned above, in the road reference line acquisition unit, every two turning points on the straight road and the trajectory points between the turning points are fitted by the least squares method, and continuous straight line segments are obtained after fitting. The angle between the direction vectors of adjacent straight line segments is calculated, and if the angle does not exceed the specified threshold, the adjacent straight line segments are merged into one segment.

[0014] This application has the following beneficial effects:

[0015] (1) This application only requires trajectory point data and laser point cloud data to realize Opendrive map construction, and has low requirements for laser radar equipment. 16-line or 32-line radars are both acceptable. The mapping process is simple, efficient and low-cost.

[0016] (2) This application superimposes the full amount of laser point cloud data onto the map, providing a comprehensive reference for map element collection and results inspection. In addition, this application is convenient and quick to collect intersection connection roads, and only the starting point and the end point are required to automatically calculate the control point, thereby obtaining a cubic Bezier curve description. The cubic Bezier curve is used to fit the curve trajectory, and the fitting result is smooth without mutations or sharp corners. The cubic Bezier curve is solved using a particle swarm intelligent optimization algorithm, which has strong algorithm solving capabilities, low computational complexity, and good fitting effect.

[0017] (3) This application only uses three-dimensional laser point cloud data to quickly supplement the information such as traffic lights and traffic signs that are missing in the tracking map and output it as a standard Opendrive map. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0019] Figure 1 is a flowchart of a mapping method provided according to an embodiment of the present application;

[0020] Figure 2 is a schematic diagram of a particle activity area provided according to an embodiment of the present application;

[0021] Figure 3is a schematic diagram of a control point position interval provided according to an embodiment of the present application;

[0022] Figure 4 is a schematic diagram of an intersection provided according to an embodiment of the present application;

[0023] Figure 5 is a schematic diagram of a method for calculating control points of a cubic Bezier curve provided in an embodiment of the present application;

[0024] Figure 6 It is a schematic diagram of the internal structure of a mapping system provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0026] This application provides a mapping method and system, which takes advantage of the low production cost of tracking maps, and combines laser point cloud data on the basis of tracking maps to quickly produce Opendrive high-precision maps containing lanes, intersections and traffic signs. After converting the tracking map into a high-precision map, the defect of the lack of detailed environmental information such as road features and traffic signs when the unmanned vehicle relies on the tracking map is solved. At the same time, the mapping process of this solution is simple and efficient, and only needs to rely on the existing tracking map and laser point cloud data. The production cost is much lower than the traditional high-precision map production process.

[0027] Embodiment 1

[0028] This application mainly combines laser point cloud data on the basis of tracking map to realize Opendrive high-precision map construction. The laser point cloud of this solution does not require the use of expensive 64-line laser radar for collection. By placing 16-line or 32-line laser radars in multiple locations, the collection coverage can be guaranteed.

[0029] like Figure 1 As shown, this embodiment provides a mapping method, which specifically includes the following steps:

[0030] Step S110: Acquire tracking map data and generate an original track.

[0031] Acquiring tracking map data includes importing the unmanned vehicle tracking map, drawing the track points in sequence, and generating the original track.

[0032] The trajectory data can be collected by the collection vehicle driving in the center of the lane.

[0033] Step S120: Acquire laser point cloud data and construct a point cloud model for display.

[0034] Acquiring laser point cloud data and constructing a point cloud model include preprocessing the acquired laser point cloud data, dividing the laser point cloud data into layers and blocks, generating a LOD (Levels of Detail) point cloud model, coloring the point cloud according to intensity, and then importing the point cloud model into the system for three-dimensional display.

[0035] The amount of point cloud data scanned by the laser radar is huge, and no system can fully load all the point cloud data. Therefore, the laser point cloud data must be lightweight first.

[0036] Export the laser point cloud to LAS format, parse the point cloud coordinates and intensity information in the LAS data, divide the laser point cloud data into blocks by grid, define the grid size of the smallest block as s, calculate the outer rectangular range of the laser point cloud data under the top-down angle, define the outer box length as h, width as w, and the layered block calculation strategy is as follows:

[0037] 1. Calculate the number of blocks in the first layer: Assume that h>w, and vice versa, the ratio of h to w is r=h / w, then the number of blocks in the first layer is n=round(r), where round means rounding to the nearest integer. The block division method is to use a grid of size w*(h / n) to divide the original rectangular outer range.

[0038] 2. Starting from the second layer, the blocks of each layer are divided into quadtrees based on the blocks of the previous layer, that is, the number of blocks in each layer is 4*n, where n is the number of blocks in the previous layer.

[0039] 3. When the division reaches the Nth layer and the length or width of the grid begins to be less than s, the division stops and the number of data layers is determined to be N-1.

[0040] 4. The number of laser point cloud data in each grid of the N-1 layer is not thinned. Starting from the N-2 layer, all the way up to the 1st layer, the point cloud in each block needs to be thinned. Assuming that the current level is k, the thinning ratio is

[0041] 5. Convert the laser point cloud data in each grid block into an osgb model node, and use the PageLod node in osg (an open source 3D rendering engine based on OpenGL) to link all blocks with a quadtree structure and output them as an osgb model.

[0042] Among them, osgb is the binary format of osg, and PageLOD is a special type of LOD node in osg.

[0043] Step S130: Obtain a road reference line based on the original trajectory and the laser point cloud data.

[0044] The key points (inflection points) in the original trajectory are extracted, and the trajectory is simplified by fitting using straight lines and cubic Bezier curves. For roads with poor trajectory fitting or no trajectory data, the simplified trajectory line is manually drawn with reference to the laser point cloud data. Finally, the simplified trajectory line is translated to the lane boundary with reference to the laser point cloud data to obtain the road reference line.

[0045] The tracking map only records discrete positioning point information. The roads on the Opendrive map need to describe the road reference line with straight lines or curve functions. Therefore, the trajectory needs to be simplified and the original trajectory needs to be fitted with straight lines and curves respectively. The specific steps include the following:

[0046] Step S1301: extract inflection points.

[0047] First, calculate the curvature of each trajectory point. For each trajectory point P, take its previous trajectory point O and the next trajectory point Q, and calculate the curvature c of point P by fitting a quadratic curve. The quadratic curve formula is defined as follows: y = a*x 2 +b*x+c, and the plane coordinates of the three points O, P, and Q are substituted into the quadratic curve formula to obtain the values ​​of coefficients a, b, and c. The first-order derivative value y'=2ax+b and the second-order derivative value y"=2a at point P are calculated, and the curvature of point P is finally solved as When the value of c is greater than the set threshold c t When , point P is the inflection point.

[0048] Step S1302: performing straight line segment fitting after inflection point extraction.

[0049] The least squares method is used to fit every two inflection points and the trajectory points between the inflection points on the straight road. For the continuous straight line segments obtained after fitting, the angle between the direction vectors of adjacent straight line segments is calculated. If the angle does not exceed the given threshold, it means that the directions of the adjacent straight line segments have not been significantly deflected and can be merged into one line segment. The least squares method is also used to fit the two adjacent straight line segments into one straight line segment.

[0050] Step S1303: performing curve segment fitting after straight line fitting.

[0051] The cubic Bezier curve is used to fit the curve trajectory between two straight line trajectories. The cubic Bezier curve is determined by 4 points, including the starting point P1, the end point P4 and the control points P2 and P3. In order to ensure the continuity of the trajectory, P1 is the end point of the previous straight line segment and P4 is the starting point of the next straight line segment. It is only necessary to determine the control points P2 and P3 to fit the cubic Bezier curve.

[0052] The coordinates of the control points P2 and P3 are solved by the particle swarm optimization algorithm (PSO), and the particle fitness is defined as the fitting error between the original trajectory point and the cubic Bezier curve. The fitting error is determined by the mean square error between the fitting point and the original point. The particle swarm algorithm formula is as follows:

[0053] v i '=ωv i +c1×rand()×(pbest i -x i )+c2×rand()×(gbest-x i ) (1)

[0054] x i '=x i +v i (2)

[0055] Formula (1) is the speed update formula, and ω is the inertia weight, which is used to control the degree of particle speed retention and help the algorithm find a balance between global search and local search. As the population iterates, the value of ω decreases linearly from 0.9 to 0.4, and pbest i represents the historical optimal position of particle i, gbest represents the historical optimal position of the population, and x i represents the current position of particle i, v i represents the current speed of particle i, c1 and c2 are learning factors, usually c1 = c2 = 2.0, and rand() represents a random weight value between [0,1]. Formula (2) is the position update formula. The new position x of the particle after each iteration is i 'From the current position x i Plus the updated speed v i 'Decide.

[0056] Define a particle as a two-dimensional point with x and y coordinates between [0,1]. Figure 2 The region class values ​​shown are as follows. The fitness is calculated by particle position as follows:

[0057] Step S13031: Determine the control point position interval.

[0058] Calculate the straight-line distance L between the starting point and the end point of the cubic Bezier curve. The optional position of the control point P2 is on the extension line of the previous straight line segment, within 2*L from the starting point. The optional position of the control point P3 is on the reverse extension line of the next straight line segment, within 2*L from the end point. The specific position range is as follows: Figure 3 shown.

[0059] Step S13032: Determine the specific coordinate positions of the control points and obtain the curve fitting results.

[0060] Assume that the x coordinate of the particle point is P x ∈[0,1], y coordinate is P y ∈[0,1].

[0061] The coordinate position of control point P2 is:

[0062] P2=P1+P x *2*L*D v1 (3)

[0063] The coordinate position of control point P3 is:

[0064] P3=P4-P y *2*L*D v4 (4)

[0065] Where D v1 and D v4 are the unit direction vectors of the previous and next straight line segments, respectively. The particle coordinates are converted to the control point positions by the above formula, and then the cubic Bezier curve equation corresponding to the control point is calculated, and then the fitness of the example is calculated by the mean square error of the trajectory fitting.

[0066] Specifically, the cubic Bezier curve is fitted by the particle swarm optimization algorithm, including the following steps:

[0067] Step P1: Initialize the particle population and randomly generate n particles within the value range.

[0068] Initialize the particle population, the number of particles is n, the number of particles should not be too large, too many particles will lead to excessive consumption of computing resources and computing time, and may cause particle congestion, but it is more likely to fall into the local extreme value and appear "premature". For the complexity of the solution of this embodiment, the number of particles can be specified as 20, that is, in Figure 2 20 points are randomly selected in the region shown. The maximum number of iterations is set to 100.

[0069] Step P2: Calculate the fitness of each initial particle, set the particle position with the best fitness as the global optimum, and set the initial position of each particle to its own historical optimal position.

[0070] Step P3: Iteratively update the particle velocity and particle position, recalculate the fitness of each particle, and update the global optimal and optimal positions of the particles.

[0071] Specifically, the position of each particle is updated using formulas (1) and (2). The particle's active area (value range) must be limited during the update. If the particle's position exceeds the active area after the update, the position update is ignored and the particle returns to its original position. When an iteration cycle ends, the fitness of each particle is recalculated, and its own historical optimal position pbest and global optimal gbest are updated.

[0072] Step P4: Determine whether the global optimal value meets the requirements or reaches the maximum number of iterations.

[0073] If the fitting error of the global optimal position calculation meets the requirement or the number of iterations reaches the limit, the current global optimal result is accepted and the fitting program is exited, and step P5 is executed; otherwise, the next iteration cycle is continued.

[0074] Step P5: Calculate the cubic Bezier curve according to the control points corresponding to the global optimum to obtain the curve fitting result.

[0075] It should be noted that the trajectories in the intersection need to be split and processed separately, without straight line or curve fitting, and collected in the subsequent intersection collection steps.

[0076] Step S1304: Connect each segment of the fitting trajectory in sequence to obtain a fitting trajectory line.

[0077] Step S1305: After translating the fitted trajectory line by a specified distance, a road reference line is obtained.

[0078] Each section of the road reference line is given a unique number, which is a positive integer.

[0079] According to the connection relationship of the reference lines, the predecessor and successor relationship of each road is determined and saved in the Opendrive map data.

[0080] The designated distance can be set in advance by the staff, and the specific value range is not limited here.

[0081] Step S140: After the road reference line is acquired, the lane data is obtained using the laser point cloud data.

[0082] Using laser point cloud data to obtain lane data includes buffering the reference line to a specified width to generate a lane. According to the point cloud information, multiple buffered lanes can be set. The lanes are automatically numbered according to the Opendrive specification and attached to the corresponding road information. At the same time, the predecessor and successor relationship of the lane is automatically generated according to the predecessor and successor relationship of the road.

[0083] Referring to the laser point cloud data, the buffered lane boundary cannot exceed the road boundary (curb), and a certain error with the boundary is allowed within the road range. The reduction in accuracy requirements can reduce the difficulty of mapping, and at the same time will not have a significant impact on the planning guidance of small unmanned vehicles.

[0084] Step S150: After the lane data is generated, intersection data collection is performed.

[0085] According to the Opendrive specification, an intersection is a place where three or more roads meet. The roads associated with it are divided into two types: incoming roads and connecting roads, such as Figure 4 shown.

[0086] In the previous steps, the incoming road collection has been completed. Now, only the connecting roads need to be collected and drawn. Since there are usually no lane lines in the intersection, only the virtual connecting road reference lines need to be collected to ensure a smooth connection with the incoming road.

[0087] In particular, the number of routes at an intersection in the tracking map may be less than 3. The system provides manual drawing capabilities, and the number of routes can be supplemented to 3 or more by referring to the laser point cloud data.

[0088] The connecting road is described by a cubic Bezier curve. The starting point and end point of the Bezier curve are determined by the endpoints of the road reference line connected at both ends. Because the connecting road is a virtual road without clear lanes, the two control points of the cubic Bezier curve can be calculated using simplified steps to simplify the collection process, such as Figure 5 As shown:

[0089] Assume that P1 and P4 are the known starting and end points of the cubic Bezier curve, and P2 and P3 are the required control points.

[0090] The method for solving the control points is: connect P1P4, take two points Q1 and Q2 on P1P4, make the lengths of P1Q1, Q1Q2 and Q2P4 equal, that is, Q1 and Q2 divide P1P4 into three equal parts, draw perpendicular lines from Q1 and Q2 to the straight lines where P1 and P4 are located, and the foot points P2 and P3 are the positions of the control points to be solved.

[0091] According to the above method, as long as the starting point and end point of the connecting road are known, the curve function description of the connecting road that is smoothly connected to the incoming road can be automatically calculated. After the connecting road reference line is generated, the lane is obtained by buffering with a fixed width, and the connection relationship between the incoming road and the connecting road in the intersection and the connection relationship between the corresponding lanes are recorded and stored in the Opendrive map data.

[0092] Step S160: After completing the intersection acquisition, the signal lights and traffic markings are collected according to the laser point cloud data, and the map is completed and exported.

[0093] With reference to the laser point cloud data, traffic signs such as traffic lights, no parking, no driving, speed bumps, speed limits, and speed limit release are marked on the map and saved to the Opendrive map data. After the required map elements are produced, the map data is exported to the xodr format, completing the mapping process from the tracking map to the Opendrive map.

[0094] Embodiment 2

[0095] like Figure 6 As shown, this embodiment provides a mapping system, which specifically includes: an original trajectory generation unit 601, a model construction unit 602, a road reference line acquisition unit 603, a lane data acquisition unit 604, an intersection acquisition unit 605, and a signal light and traffic marking acquisition unit 606.

[0096] The original trajectory generating unit 601 is used to obtain the tracking map data and generate the original trajectory.

[0097] Acquiring tracking map data includes importing the unmanned vehicle tracking map, drawing the track points in sequence, and generating the original track.

[0098] The trajectory data can be collected by the collection vehicle driving in the center of the lane.

[0099] The model building unit 602 is used to acquire laser point cloud data and build a point cloud model for display.

[0100] Acquiring laser point cloud data and constructing a point cloud model include preprocessing the acquired laser point cloud data, dividing the laser point cloud data into layers and blocks, generating a LOD (Levels of Detail) point cloud model, coloring the point cloud according to intensity, and then importing the point cloud model into the system for three-dimensional display.

[0101] The amount of point cloud data scanned by the laser radar is huge, and no system can fully load all the point cloud data. Therefore, the laser point cloud data must be lightweight first.

[0102] Export the laser point cloud to LAS format, parse the point cloud coordinates and intensity information in the LAS data, divide the laser point cloud data into blocks by grid, define the grid size of the smallest block as s, calculate the outer rectangular range of the laser point cloud data under the top-down angle, define the outer box length as h, width as w, and the layered block calculation strategy is as follows:

[0103] 1. Calculate the number of blocks in the first layer: Assume that h>w, and vice versa, the ratio of h to w is r=h / w, then the number of blocks in the first layer is n=round(r), where round means rounding to the nearest integer. The block division method is to use a grid of size w*(h / n) to divide the original rectangular outer range.

[0104] 2. Starting from the second layer, the blocks of each layer are divided into quadtrees based on the blocks of the previous layer, that is, the number of blocks in each layer is 4*n, where n is the number of blocks in the previous layer.

[0105] 3. When the division reaches the Nth layer and the length or width of the grid begins to be less than s, the division stops and the number of data layers is determined to be N-1.

[0106] 4. The number of laser point cloud data in each grid of the N-1 layer is not thinned. Starting from the N-2 layer, all the way up to the 1st layer, the point cloud in each block needs to be thinned. Assuming that the current level is k, the thinning ratio is

[0107] 5. Convert the laser point cloud data in each grid block into an osgb model node, and use the PageLod node in osg (an open source 3D rendering engine based on OpenGL) to link all blocks with a quadtree structure and output them as an osgb model.

[0108] Among them, osgb is the binary format of osg, and PageLOD is a special type of LOD node in osg.

[0109] The road reference line acquisition unit 603 is used to acquire the road reference line according to the original trajectory and the laser point cloud data.

[0110] The key points (inflection points) in the original trajectory are extracted, and the trajectory is simplified by fitting using straight lines and cubic Bezier curves. For roads with poor trajectory fitting or no trajectory data, the simplified trajectory line is manually drawn with reference to the laser point cloud data. Finally, the simplified trajectory line is translated to the lane boundary with reference to the laser point cloud data to obtain the road reference line.

[0111] The tracking map only records discrete positioning point information. The roads on the Opendrive map need to describe the road reference line with straight lines or curve functions. Therefore, the trajectory needs to be simplified and the original trajectory needs to be fitted with straight lines and curves respectively. The specific steps include the following:

[0112] Step W1: extract inflection points.

[0113] First, calculate the curvature of each trajectory point. For each trajectory point P, take its previous trajectory point O and the next trajectory point Q, and calculate the curvature c of point P by fitting a quadratic curve. The quadratic curve formula is defined as follows: y = a*x 2 +b*x+c, and the plane coordinates of the three points O, P, and Q are substituted into the quadratic curve formula to obtain the values ​​of coefficients a, b, and c. The first-order derivative value y'=2ax+b and the second-order derivative value y"=2a at point P are calculated, and the curvature of point P is finally solved as When the value of c is greater than the set threshold c t When , point P is the inflection point.

[0114] Step W2: Perform straight line segment fitting after inflection point extraction.

[0115] The least squares method is used to fit every two inflection points and the trajectory points between the inflection points on the straight road. For the continuous straight line segments obtained after fitting, the angle between the direction vectors of adjacent straight line segments is calculated. If the angle does not exceed the given threshold, it means that the directions of the adjacent straight line segments have not been significantly deflected and can be merged into one line segment. The least squares method is also used to fit the two adjacent straight line segments into one straight line segment.

[0116] Step W3: After straight line fitting, curve segment fitting is performed.

[0117] The cubic Bezier curve is used to fit the curve trajectory between two straight line trajectories. The cubic Bezier curve is determined by 4 points, including the starting point P1, the end point P4 and the control points P2 and P3. In order to ensure the continuity of the trajectory, P1 is the end point of the previous straight line segment and P4 is the starting point of the next straight line segment. It is only necessary to determine the control points P2 and P3 to fit the cubic Bezier curve.

[0118] The coordinates of the control points P2 and P3 are solved by the particle swarm optimization algorithm (PSO), and the particle fitness is defined as the fitting error between the original trajectory point and the cubic Bezier curve. The fitting error is determined by the mean square error between the fitting point and the original point. The particle swarm algorithm formula is as follows:

[0119] v i '=ωv i +c1×rand()×(pbest i -x i )+c2×rand()×(gbest-x i ) (1)

[0120] x i '=x i +v i (2)

[0121] Formula (1) is the speed update formula, and ω is the inertia weight, which is used to control the degree of particle speed retention and help the algorithm find a balance between global search and local search. As the population iterates, the value of ω decreases linearly from 0.9 to 0.4, and pbest i represents the historical optimal position of particle i, gbest represents the historical optimal position of the population, and x i represents the current position of particle i, v irepresents the current speed of particle i, c1 and c2 are learning factors, usually c1 = c2 = 2.0, and rand() represents a random weight value between [0,1]. Formula (2) is the position update formula. The new position x of the particle after each iteration is i 'From the current position x i Plus the updated speed v i 'Decide.

[0122] Define a particle as a two-dimensional point with x and y coordinates between [0,1]. Figure 2 The region class values ​​shown are as follows. The fitness is calculated by particle position as follows:

[0123] Step W31: Determine the control point position interval.

[0124] Calculate the straight-line distance L between the starting point and the end point of the cubic Bezier curve. The optional position of the control point P2 is on the extension line of the previous straight line segment, within 2*L from the starting point. The optional position of the control point P3 is on the reverse extension line of the next straight line segment, within 2*L from the end point. The specific position range is as follows: Figure 3 shown.

[0125] Step W32: Determine the specific coordinate positions of the control points and obtain the curve fitting results.

[0126] Assume that the x coordinate of the particle point is P x ∈[0,1], y coordinate is P y ∈[0,1].

[0127] The coordinate position of control point P2 is:

[0128] P2=P1+P x *2*L*D v1 (3)

[0129] The coordinate position of control point P3 is:

[0130] P3=P4-P y *2*L*D v4 (4)

[0131] Where D v1 and D v4 are the unit direction vectors of the previous and next straight line segments, respectively. The particle coordinates are converted to the control point positions by the above formula, and then the cubic Bezier curve equation corresponding to the control point is calculated, and then the fitness of the example is calculated by the mean square error of the trajectory fitting.

[0132] Specifically, the cubic Bezier curve is fitted by the particle swarm optimization algorithm, including the following steps:

[0133] Step P1: Initialize the particle population and randomly generate n particles within the value range.

[0134] Initialize the particle population, the number of particles is n, the number of particles should not be too large, too many particles will lead to excessive consumption of computing resources and computing time, and may cause particle congestion, but it is more likely to fall into the local extreme value and appear "premature". For the complexity of the solution of this embodiment, the number of particles can be specified as 20, that is, in Figure 2 20 points are randomly selected in the region shown. The maximum number of iterations is set to 100.

[0135] Step P2: Calculate the fitness of each initial particle, set the particle position with the best fitness as the global optimum, and set the initial position of each particle to its own historical optimal position.

[0136] Step P3: Iteratively update the particle velocity and particle position, recalculate the fitness of each particle, and update the global optimal and optimal positions of the particles.

[0137] Specifically, the position of each particle is updated using formulas (1) and (2). The particle's active area (value range) must be limited during the update. If the particle's position exceeds the active area after the update, the position update is ignored and the particle returns to its original position. When an iteration cycle ends, the fitness of each particle is recalculated, and its own historical optimal position pbest and global optimal gbest are updated.

[0138] Step P4: Determine whether the global optimal value meets the requirements or reaches the maximum number of iterations.

[0139] If the fitting error of the global optimal position calculation meets the requirement or the number of iterations reaches the limit, the current global optimal result is accepted and the fitting program is exited, and step P5 is executed; otherwise, the next iteration cycle is continued.

[0140] Step P5: Calculate the cubic Bezier curve according to the control points corresponding to the global optimum to obtain the curve fitting result.

[0141] It should be noted that the trajectories in the intersection need to be split and processed separately, without straight line or curve fitting, and collected in the subsequent intersection collection steps.

[0142] Step W4: Connect each segment of the fitting trajectory in sequence to obtain a fitting trajectory line.

[0143] Step W5: After translating the fitted trajectory line by a specified distance, a road reference line is obtained.

[0144] Each section of the road reference line is given a unique number, which is a positive integer.

[0145] According to the connection relationship of the reference lines, the predecessor and successor relationship of each road is determined and saved in the Opendrive map data.

[0146] The designated distance can be set in advance by the staff, and the specific value range is not limited here.

[0147] The lane data acquisition unit 604 is used to obtain lane data using laser point cloud data.

[0148] Using laser point cloud data to obtain lane data includes buffering the reference line to a specified width to generate a lane. According to the point cloud information, multiple buffered lanes can be set. The lanes are automatically numbered according to the Opendrive specification and attached to the corresponding road information. At the same time, the predecessor and successor relationship of the lane is automatically generated according to the predecessor and successor relationship of the road.

[0149] Referring to the laser point cloud data, the buffered lane boundary cannot exceed the road boundary (curb), and a certain error with the boundary is allowed within the road range. The reduction in accuracy requirements can reduce the difficulty of mapping, and at the same time will not have a significant impact on the planning guidance of small unmanned vehicles.

[0150] The intersection collection unit 605 is used to collect intersection data.

[0151] According to the Opendrive specification, an intersection is a place where three or more roads meet. The roads associated with it are divided into two types: incoming roads and connecting roads, such as Figure 4 shown.

[0152] In the previous steps, the incoming road collection has been completed. Now, only the connecting roads need to be collected and drawn. Since there are usually no lane lines in the intersection, only the virtual connecting road reference lines need to be collected to ensure a smooth connection with the incoming road.

[0153] In particular, the number of routes at an intersection in the tracking map may be less than 3. The system provides manual drawing capabilities, and the number of routes can be supplemented to 3 or more by referring to the laser point cloud data.

[0154] The connecting road is described by a cubic Bezier curve. The starting point and end point of the Bezier curve are determined by the endpoints of the road reference line connected at both ends. Because the connecting road is a virtual road without clear lanes, the two control points of the cubic Bezier curve can be calculated using simplified steps to simplify the collection process, such as Figure 5 As shown:

[0155] Assume that P1 and P4 are the known starting and end points of the cubic Bezier curve, and P2 and P3 are the required control points.

[0156] The method for solving the control points is: connect P1P4, take two points Q1 and Q2 on P1P4, make the lengths of P1Q1, Q1Q2 and Q2P4 equal, that is, Q1 and Q2 divide P1P4 into three equal parts, draw perpendicular lines from Q1 and Q2 to the straight lines where P1 and P4 are located, and the foot points P2 and P3 are the positions of the control points to be solved.

[0157] According to the above method, as long as the starting point and end point of the connecting road are known, the curve function description of the connecting road that is smoothly connected to the incoming road can be automatically calculated. After the connecting road reference line is generated, the lane is obtained by buffering with a fixed width, and the connection relationship between the incoming road and the connecting road in the intersection and the connection relationship between the corresponding lanes are recorded and stored in the Opendrive map data.

[0158] The signal light and traffic marking collection unit 606 is used to collect signal lights and traffic markings according to the laser point cloud data, and complete and export the map.

[0159] With reference to the 3D laser point cloud data, traffic signs such as traffic lights, no parking, no driving, speed bumps, speed limits, and speed limit release are marked on the map and saved to the Opendrive map data. After the required map elements are produced, the map data is exported to the xodr format, completing the mapping process from the tracking map to the Opendrive map.

[0160] This application has the following beneficial effects:

[0161] (1) This application only requires trajectory point data and laser point cloud data to realize Opendrive map construction, and has low requirements for laser radar equipment. 16-line or 32-line radars are both acceptable. The mapping process is simple, efficient and low-cost.

[0162] (2) This application superimposes the full amount of laser point cloud data onto the map, providing a comprehensive reference for map element collection and results inspection. In addition, this application is convenient and quick to collect intersection connection roads, and only the starting point and the end point are required to automatically calculate the control point, thereby obtaining a cubic Bezier curve description. The cubic Bezier curve is used to fit the curve trajectory, and the fitting result is smooth without mutations or sharp corners. The cubic Bezier curve is solved using a particle swarm intelligent optimization algorithm, which has strong algorithm solving capabilities, low computational complexity, and good fitting effect.

[0163] (3) This application only uses three-dimensional laser point cloud data to quickly supplement the information such as traffic lights and traffic signs that are missing in the tracking map and output it as a standard Opendrive map.

[0164] Although the present application is described with reference to examples, this is for illustrative purposes only and is not intended to limit the present application, and changes, additions and / or deletions to the embodiments may be made without departing from the scope of the present application.

[0165] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A mapping method, characterized in that: The following steps are involved: Obtain the tracking map data and generate the original trajectory; Acquire laser point cloud data and build a point cloud model for display; Obtain road reference lines based on original trajectory and laser point cloud data; After obtaining the road reference line, the lane data is obtained using the laser point cloud data; After generating lane data, conduct intersection collection; After completing the intersection collection, the signal lights and traffic markings are collected based on the laser point cloud data, and the map is completed and exported.

2. The mapping method according to claim 1, characterized in that: Acquiring tracking map data includes importing the unmanned vehicle tracking map, drawing the track points in sequence, and generating the original track.

3. The mapping method according to claim 1, wherein: Based on the original trajectory and laser point cloud data, obtaining the road reference line includes the following sub-steps: Extract inflection points; After the inflection point is extracted, straight line segment fitting is performed; After straight line fitting, curve segment fitting is performed; Connect each segment of the fitting trajectory in turn to obtain the fitting trajectory line; After translating the fitted trajectory line by a specified distance, the road reference line is obtained.

4. The mapping method according to claim 3, characterized in that: The inflection point extraction includes determining the curvature of each trajectory point, where the curvature c is expressed as: The first-order derivative value y'=2ax+b, the second-order derivative value y"=2a, a and b are derivative coefficients; When the curvature c is greater than the specified threshold, the current trajectory point is an inflection point.

5. The mapping method according to claim 3, characterized in that: Straight line segment fitting includes performing least squares straight line fitting on every two inflection points on the straight road and the trajectory points between the inflection points, obtaining continuous straight line segments after fitting, calculating the angle between the direction vectors of adjacent straight line segments, and merging the adjacent straight line segments into one segment if the angle does not exceed a specified threshold.

6. A mapping system, characterized in that: Specifically include: Original trajectory generation unit, model building unit, road reference line acquisition unit, lane data acquisition unit, intersection acquisition unit, and signal light and traffic marking acquisition unit; An original trajectory generating unit, used to obtain tracking map data and generate an original trajectory; A model building unit, used to acquire laser point cloud data and build a point cloud model for display; A road reference line acquisition unit, used for acquiring a road reference line according to the original trajectory and the laser point cloud data; A lane data acquisition unit, used for acquiring lane data using laser point cloud data; An intersection collection unit, used for collecting intersection data; The signal light and traffic marking acquisition unit is used to collect signal lights and traffic markings based on laser point cloud data, complete and export maps.

7. The mapping system according to claim 6, wherein: The original trajectory generation unit obtains the tracking map data by importing the unmanned vehicle tracking map and drawing the trajectory points in sequence to generate the original trajectory.

8. The mapping system according to claim 7, wherein: The road reference line acquisition unit acquires the road reference line according to the original trajectory and the laser point cloud data, including the following sub-steps: Extract inflection points; After the inflection point is extracted, straight line segment fitting is performed; After straight line fitting, curve segment fitting is performed; Connect each segment of the fitting trajectory in turn to obtain the fitting trajectory line; After translating the fitted trajectory line by a specified distance, the road reference line is obtained.

9. The mapping system according to claim 8, characterized in that: The inflection point extraction in the road reference line acquisition unit includes determining the curvature of each trajectory point, where the curvature c is expressed as: The first-order derivative value y'=2ax+b, the second-order derivative value y"=2a, a and b are derivative coefficients; When the curvature c is greater than the specified threshold, the current trajectory point is an inflection point.

10. The mapping system according to claim 8, wherein: In the road reference line acquisition unit, the least squares method is used to fit every two inflection points on the straight road and the trajectory points between the inflection points. After fitting, continuous straight line segments are obtained, and the angle between the direction vectors of adjacent straight line segments is calculated. If the angle does not exceed the specified threshold, the adjacent straight line segments are merged into one segment.

Citation Information

Cited By

  • High-adaptability fruit tea garden equipment autonomous navigation method based on low-precision map

    CN121207208A