A method for generating 3D high-precision vector maps based on trajectory information

Abstract

The present invention provides a method for generating a 3D high-precision vector map based on trajectory information. First, a 2D vector map is drawn based on the current mature 2D vector mapping technology, and then the 2D map is automatically converted into a 3D map based on the vehicle trajectory information. By avoiding the complex and time-consuming 3D recognition and 3D human-computer interactive mapping links, the mapping efficiency is improved.

Description

Technical Field

[0001] The invention relates to a method for generating a 3D high-precision vector map based on trajectory information. Background Art

[0002] High-precision vector maps are special maps for autonomous driving. By modeling the driving environment, high-precision vector maps can provide important support for vehicle positioning, perception, path planning and decision-making. Figure 1 As shown in the figure, the core content of the vector map is the spatial topological association and connection relationship between geometric entities such as lane center lines, lane edges, stop lines, traffic lights, and crosswalks, as well as between these entities.

[0003] According to the map requirements of unmanned vehicles, existing high-precision vector maps can be divided into two specifications: 2D vector maps and 3D vector maps. Among them, 2D maps assume that the ground is a horizontal plane, and obtain the map representation of X and Y dimensions by orthographic projection of real data along the Z axis. Based on the 2D map, the real elevation of each ground position is modeled to obtain a 3D map represented in three dimensions of X, Y, and Z.

[0004] 3D vector maps have always been produced mainly based on 3D laser point clouds. In the early days of high-precision map development, the mapping process was mainly based on a fully manual 3D interactive system. In order to obtain detailed 3D road vector information, a lot of manpower and material resources are required for map drawing. In recent years, with the widespread application of artificial intelligence and deep learning technologies, the use of automatic recognition algorithms for initial recognition and drawing of map elements, and the use of 3D interactive systems for manual editing and verification based on the initial drawing are becoming the mainstream mapping technology at this stage.

[0005] Although the production efficiency of 3D vector maps can be improved to a certain extent through automatic recognition and vectorization algorithms, due to the complexity of road scenes and the low fault tolerance requirements of autonomous vehicles for map data, the manual editing and verification of 3D vector maps still requires a lot of manual participation for existing technologies. Low mapping efficiency, high mapping costs, and long mapping cycles have become obstacles to the large-scale application of high-precision maps. In addition, in order to obtain reliable 3D recognition results, a lot of extra effort is required for sample data collection and manual annotation, which further increases the cost of map production. Summary of the invention

[0006] The technical problem to be solved by the present invention is: in order to overcome the deficiencies in the prior art, the present invention provides a 3D high-precision vector map generation method based on trajectory information, which can produce 3D vector maps quickly, efficiently and at low cost.

[0007] The technical solution to be adopted by the present invention to solve the technical problem is: a method for generating a 3D high-precision vector map based on trajectory information, comprising the following steps:

[0008] S1: 2D projection and 2D vector graphics production

[0009] Acquire 3D point cloud data, perform orthographic projection on the 3D point cloud data to reduce its dimension into 2D images, automatically identify and manually edit map elements based on the projected 2D images, and then use 2D projection and mapping technology to process the 2D images to obtain a 2D high-precision vector map of uniform specifications;

[0010] S2: Ground elevation calculation

[0011] Based on the map, the driving trajectory information recorded by the vehicle and the vehicle height information are collected to calculate the 3D coordinates of the ground point at the corresponding position of the driving trajectory;

[0012] S3: Keypoint extraction

[0013] Formulate a key point screening strategy, and downsample the dense trajectory points (i.e., all trajectory points) obtained in step S2 according to the key point screening strategy, and then screen out local key points as representatives of ground elevation values ​​to form a trajectory ground key point set, where downsampling refers to the process of screening key points from a large number of original points; this step integrates the road scene elevation prior information and the trajectory point spatial distribution information, and proposes three effective key point screening criteria. The combination of the three criteria can accurately express the true elevation of the road surface with fewer key points, thereby improving the efficiency and quality of subsequent 3D conversion.

[0014] S4: 3D Conversion

[0015] According to the trajectory ground key point set obtained in step S3, the elevation values ​​corresponding to all nodes in the 2D vector map are calculated point by point by weighted interpolation. A 3D vector map is formed, wherein nodes refer to all coordinate points stored in the vector map. This step proposes an elevation interpolation method based on key point distance weighting, and its technical effect is that it can achieve higher interpolation accuracy at a lower computational cost.

[0016] Specifically, the driving trajectory information is a time-series 3D coordinate point string output by the integrated navigation device carried by the map collection vehicle, and the vehicle height information is a vertical distance from the integrated navigation device carried by the map collection vehicle to the ground, denoted as h.

[0017] Specifically, the process of calculating the 3D coordinates of the ground point in step S2 is as follows:

[0018] The target trajectory point P on the known driving trajectory i The coordinates of i (xi ,y i ,z i ), target trajectory point P i The nearest reference trajectory point P j The coordinates of j (x j ,y j ,z j ), then: target trajectory point P i The corresponding ground point P i * The coordinates P i * (x i * ,y i * ,z i * ) is calculated as:

[0019]

[0020]

[0021]

[0022]

[0023] Where h is the vehicle height and θ is the target trajectory point P i The approximate slope of the ground at the position, the subscripts i and j represent the trajectory point numbers, i = 1, 2, 3...n, j = 1, 2, 3...n, and i ≠ j.

[0024] In step S2, ground slope information is introduced into the elevation solution process, so that higher solution accuracy can be achieved for specific road scenes such as up and down ramps.

[0025] Specifically, the key point screening strategy in step S3 includes the following three criteria:

[0026] 1) If the current trajectory point P i The first or last point of the trajectory P is recorded as a key point, as shown in formula (5):

[0027] i=1||n=size(P) (5)

[0028] Wherein, 1 represents the sequence number of the first point of the track P, and n represents the sequence number of the last point of the track P, that is, n = size(P);

[0029] 2) If the current trajectory point P i With the previous key point P k The length of the trajectory passed by is greater than the threshold th pathThen it is recorded as the key point, as shown in formula (6):

[0030] ||path(P k ,P i )||>th path (6)

[0031] Threshold path Determined based on experience.

[0032] 3) If the current trajectory point P i If it is a local elevation extreme point, it is recorded as a key point, as shown in formula (7):

[0033] z i =max(z k )||z i =min(z k )(k∈[i-ε,i+ε]) (7)

[0034] Among them, size(P) is the total number of points contained in the trajectory, path(P k ,P i ) is the measurement function of the trajectory length between two points, ε is the neighborhood radius, and the local elevation extreme point specifically refers to the local maximum and minimum values ​​of the elevation value.

[0035] Specifically, the calculation of the node elevation value in step S4 is as follows:

[0036] For the node P recorded in the 2D vector map 0 , whose coordinates are marked as P 0 (x 0 ,y 0 ), first according to the 2D coordinate P of the node 0 (x 0 ,y 0 ), search for node P in the trajectory ground key point set 0 The k nearest neighbors of , and the weight w of each neighbor point is calculated according to formula (8) i :

[0037]

[0038]

[0039] Among them, d i Represents the distance between two points; x i ,y i Represents the coordinate value of the adjacent point; k nearest neighbor refers to the distance P 0 The k points closest to a point.

[0040] Then, the elevation value of the 2D node is obtained by weighted summing the elevations of all k nearest neighbor points. As shown in formula (9):

[0041]

[0042] All nodes in the 2D vector map are traversed in sequence and their elevation values ​​are solved according to the above steps, and finally the converted 3D vector map can be obtained.

[0043] The beneficial effects of the present invention are as follows: the present invention provides a method for generating a 3D high-precision vector map based on trajectory information, firstly draws a 2D vector map based on the current mature 2D vector mapping technology, and then automatically converts the 2D map into a 3D map based on the vehicle trajectory information. By avoiding the complex and time-consuming 3D recognition and 3D human-computer interaction mapping links, the mapping efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0045] Figure 1 It is a schematic diagram of a high-precision vector map in the prior art.

[0046] Figure 2 It is a technical flow chart of the method for generating a 3D high-precision vector map based on trajectory information of the present invention. DETAILED DESCRIPTION

[0047] The present invention will now be described in detail with reference to the accompanying drawings. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, and therefore only shows the components related to the present invention.

[0048] like Figure 2 As shown, a 3D high-precision vector map generation method based on trajectory information of the present invention takes 3D point cloud data, driving trajectory information and vehicle height information obtained by a map collection vehicle as input, and finally obtains a "3D high-precision vector map" through four processes of "2D projection and 2D vector map production", "ground elevation solution", "key point extraction" and "3D conversion".

[0049] The implementation process of each step in the technical flow chart will be explained in detail below.

[0050] Step S1: 2D projection and 2D vector graphics production

[0051] Compared with 3D point clouds, the automated processing methods for 2D images are more mature, and the amount of data is greatly reduced after dimensionality reduction, and the processing speed is more advantageous. In addition, for existing human-computer interaction systems, the application threshold and complexity of 3D interaction are significantly higher than those of 2D interaction. Therefore, in the process of generating high-precision maps (especially 2D high-precision vector maps), ortho-projecting 3D data into 2D images, and automatically identifying and manually editing map elements based on the projected 2D images have been proven by the industry to significantly improve mapping efficiency and reduce mapping costs and complexity.

[0052] In this step, the data is processed using various publicly available 2D projection and mapping technologies, including but not limited to, to ultimately obtain a 2D high-precision vector map of uniform specifications, which is then elevated to 3D space based on subsequent steps.

[0053] Step S2: Ground elevation calculation

[0054] This step calculates the 3D coordinates of the ground point at the corresponding position of the trajectory based on the trajectory information recorded by the map acquisition vehicle and the vehicle height information. The trajectory information is the time-series 3D coordinate point string output by the integrated navigation device carried by the map acquisition vehicle, and the vehicle height information is the vertical distance from the integrated navigation device to the ground, recorded as h.

[0055] The target trajectory point P on the known driving trajectory i The coordinates of (current trajectory point) are P i (x i ,y i ,z i ), target trajectory point P i The nearest reference trajectory point P j The coordinates of j (x j ,y j ,z j ), target trajectory point P i The corresponding ground point is denoted as P i * , calculate the 3D coordinates of the ground point corresponding to each trajectory point, and the neighboring reference trajectory points are used to assist in calculating the target trajectory point P i The ground at location has an approximate slope of θ.

[0056] Then: The target trajectory point P i The corresponding ground point P i * The coordinates P i * (x i * ,y i * ,z i *) is calculated as:

[0057]

[0058]

[0059]

[0060]

[0061] Where h is the vehicle height and θ is the target trajectory point P i The approximate slope of the ground at the position, the subscripts i and j represent the trajectory point numbers, i = 1, 2, 3...n, j = 1, 2, 3...n, and i ≠ j.

[0062] Step S3: Key point extraction

[0063] In this step, the dense trajectory points obtained in the previous step are downsampled by formulating a key point screening strategy, and then local key points are screened out as representatives of ground elevation values ​​to form a trajectory ground key point set to serve the subsequent 3D interpolation conversion.

[0064] Specifically, key points are extracted according to the following three criteria:

[0065] 1) If the current trajectory point P i The first or last point of the trajectory P is recorded as a key point, as shown in formula (5):

[0066] i=1||n=size(P) (5)

[0067] Among them, 1 represents the sequence number of the first point of track P, and n represents the sequence number of the last point of track P.

[0068] 2) If the current trajectory point P i With the previous key point P k The length of the trajectory passed by is greater than the threshold th path Then it is recorded as the key point, as shown in formula (6):

[0069] ||path(P k ,P i )||>th path (6)

[0070] 3) If the current trajectory point P i If it is a local elevation extreme point, it is recorded as a key point, as shown in formula (7):

[0071] z i =max(z k )||z i =min(z k)(k∈[i-ε,i+ε]) (7)

[0072] Among them, size(P) is the total number of points contained in the trajectory, path(P k ,P i ) is the trajectory length measurement function between two points, and ε is the neighborhood radius.

[0073] Step S4: 3D Conversion

[0074] This step mainly calculates the elevation values ​​corresponding to all nodes in the 2D vector map point by point through weighted interpolation based on the trajectory ground key point set obtained in step S3. Form a 3D vector map.

[0075] Specifically, the present invention is directed to a node P recorded in a 2D vector map. 0 , whose coordinates are marked as P 0 (x 0 ,y 0 ), first according to the 2D coordinate P of the node 0 (x 0 ,y 0 ), search for node P in the trajectory ground key point set 0 The k nearest neighbors of , and the weight w of each neighbor point is calculated according to formula (8) i :

[0076]

[0077]

[0078] Among them, d i Represents the distance between two points; x i ,y i Indicates the coordinate value of the nearby point;

[0079] Then, the elevation value of the 2D node is obtained by weighted summing the elevations of all k nearest neighbor points. As shown in formula (9):

[0080]

[0081] Finally, all nodes in the 2D vector map are traversed in turn and their elevation values ​​are solved according to the above steps, and finally the converted 3D vector map can be obtained.

[0082] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the scope of the present invention through the above description. The technical scope of this invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for generating a 3D high-precision vector map based on trajectory information, characterized in that: The following steps are involved: S1: 2D projection and 2D vector graphics production Acquire 3D point cloud data, perform orthographic projection on the 3D point cloud data to reduce its dimension into 2D images, automatically identify and manually edit map elements based on the projected 2D images, and then use 2D projection and mapping technology to process the 2D images to obtain a 2D high-precision vector map of uniform specifications; S2: Ground elevation calculation Based on the map, the driving trajectory information recorded by the vehicle and the vehicle height information are collected to calculate the 3D coordinates of the ground point at the corresponding position of the driving trajectory; S3: Keypoint extraction Formulate a key point screening strategy, downsample the dense trajectory points obtained in step S2 according to the key point screening strategy, and then screen out local key points as representatives of ground elevation values ​​to form a trajectory ground key point set; S4: 3D Conversion According to the trajectory ground key point set obtained in step S3, the elevation values ​​corresponding to all nodes in the 2D vector map are calculated point by point by weighted interpolation. Form a 3D vector map.

2. The method for generating a 3D high-precision vector map based on trajectory information according to claim 1, characterized in that: The driving trajectory information is a time-series 3D coordinate point string output by the integrated navigation device carried by the map collection vehicle, and the vehicle height information is a vertical distance from the integrated navigation device carried by the map collection vehicle to the ground, denoted as h.

3. The method for generating a 3D high-precision vector map based on trajectory information according to claim 2, characterized in that: The process of calculating the 3D coordinates of the ground point in step S2 is: The target trajectory point P on the known driving trajectory i The coordinates of i (x i ,y i ,z i ), target trajectory point P i The nearest reference trajectory point P j The coordinates of j (x j ,y j ,z j ), then: target trajectory point P i The corresponding ground point P i * The coordinates P i * (x i * ,y i * ,z i * ) is calculated as: Where h is the vehicle height and θ is the target trajectory point P i The approximate slope of the ground at the position, the subscripts i and j represent the trajectory point numbers, i = 1, 2, 3...n, j = 1, 2, 3...n, and i ≠ j.

4. The method for generating a 3D high-precision vector map based on trajectory information according to claim 1, characterized in that: The key point screening strategy in step S3 includes the following three criteria: 1) If the current trajectory point P i The first or last point of the trajectory P is recorded as a key point, as shown in formula (5): i=1||n (5) Wherein, 1 represents the sequence number of the first point of the track P, and n represents the sequence number of the last point of the track P, that is, n = size(P); 2) If the current trajectory point P i With the previous key point P k The trajectory length between path Then it is recorded as the key point, as shown in formula (6): ||path(P k ,P i )||>th path (6) 3) If the current trajectory point P i If it is a local elevation extreme point, it is recorded as a key point, as shown in formula (7): With i =max(from k )||from i =min(of k )(k∈[i-ε,i+ε]) (7) Among them, size(P) is the total number of points contained in the trajectory, path(P k ,P i ) is the trajectory length measurement function between two points, and ε is the neighborhood radius.

5. The method for generating a 3D high-precision vector map based on trajectory information according to claim 1, characterized in that: The calculation of the node elevation value in step S4 is specifically as follows: For the node P0 recorded in the 2D vector map, whose coordinates are marked as P0(x0,y0), firstly, according to the 2D coordinates of the node P0(x0,y0), the k nearest neighbors of the node P0 are searched in the trajectory ground key point set, and the weight w of each neighbor point is calculated according to formula (8): i : Among them, d i Indicates the distance between two points; x i ,y i Indicates the coordinate value of the nearby point; Then, the elevation value of the 2D node is obtained by weighted summing the elevations of all k nearest neighbor points. As shown in formula (9): All nodes in the 2D vector map are traversed in sequence and their elevation values ​​are solved according to the above steps, and finally the converted 3D vector map can be obtained.

Images