Unmanned vehicle collision detection method based on seven-circle method

Abstract

The invention discloses an unmanned vehicle collision detection method based on a seven-circle method. The method comprises the following steps: 1) establishing local grid maps according to a laser radar and a camera; 2) traversing all grids in the local grid map, calculating the distance from each grid to the nearest barrier grid, and storing the distance in the grid; 3) projecting the unmanned vehicle into the local grid map, and establishing a seven-circle collision model on the unmanned vehicle projected into the local grid map; 4) carrying out collision inspection; taking the path discrete point as the projection center point of the unmanned vehicle, if the distance to the nearest obstacle of the unmanned vehicle is greater than the radius of the large circumcircle, the unmanned vehicle does not collide, and if the distance to the nearest obstacle of the unmanned vehicle is less than or equal to the radius of the large circumcircle, the unmanned vehicle does not collide. If yes, the distance from the obstacle closest to the unmanned vehicle to the radiuses of all the middle circumcircle and the small circumcircle is judged, if the distance from the obstacle closest to the unmanned vehicle to the radiuses of all the middle circumcircle and the small circumcircle is larger than the radiuses of all the middle circumcircle and the small circumcircle, the unmanned vehicle cannot collide with the obstacle, and otherwise, the unmanned vehicle collides with the obstacle.

Description

Technical Field

[0001] The invention relates to the field of vehicle collision detection, and in particular to a vehicle collision detection method based on a seven-circle method. Background Art

[0002] In order to ensure the safe driving of driverless vehicles, driverless vehicles will continuously perform collision detection when driving on the path to ensure that the driverless vehicles will not collide with obstacles on the way. When planning the path, the driverless vehicle will discretize the path into waypoints one by one for storage. To perform collision detection on the path, it is necessary to perform collision detection on all discrete points on the path.

[0003] The selection of general collision detection methods is often closely related to the map and obstacle data structure in unmanned driving. Among the existing collision detection methods, raster traversal method and vector method are commonly used.

[0004] Grid traversal method: First, the entire continuous space is divided into discrete, equal-volume small grids to form a local grid map; then the obstacles are projected in the discrete space, and the grid occupancy is marked on the local grid map after projection; finally, by judging whether each grid occupied by the vehicle border model in the grid map is occupied by an obstacle, it is judged whether the unmanned vehicle has collided. The accuracy of this method depends on the grid resolution, and when the vehicle heading is not vertical or horizontal, additional calculations are required to determine the grid area covered by the vehicle border, and when the vehicle size increases, the number of traversed grids also increases. Therefore, this method will greatly increase the computational burden of the system when performing grid traversal at each planned path point. For example, a vehicle with a length of 5 meters and a width of 2 meters occupies 250 grids in a grid map with a resolution of 0.2 meters. When the path length is 50 meters and the waypoint interval is 0.5 meters, the total number of grids that need to be traversed for the entire path is 25,000. It can be seen that grid traversal will consume a considerable amount of time, making collision detection unable to meet the real-time requirements.

[0005] Vector method: Encapsulate the perceived obstacle into an outline to form a bounding box. Then compare the vector graphics formed by the vehicle collision frame with the bounding box one by one, and determine whether there is a collision by judging whether the two overlap. This method requires abstracting the obstacle into a bounding box. For square objects, the bounding box can well reflect its real space occupancy in space, but for irregular objects, especially those with concave and convex shapes, the bounding box cannot reflect its real space occupancy, resulting in some passable areas being incorrectly assigned obstacle information, thereby reducing the accuracy of collision detection. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention provides an unmanned vehicle collision detection method based on the seven-circle method, which has small calculation amount, high detection accuracy and strong real-time performance.

[0007] The technical solution of the present invention is: a collision detection method for an unmanned vehicle based on the seven-circle method, comprising the following steps:

[0008] 1) Establish a local grid map of the environment around the unmanned vehicle based on the laser radar and camera respectively;

[0009] 2) Traverse all grids in the local grid map, calculate the distance between each grid and its nearest obstacle grid, and store the distance in the grid;

[0010] 3) Projecting the unmanned vehicle into the local grid map, and establishing a seven-circle collision model on the unmanned vehicle projected into the local grid map, wherein the steps for establishing the seven-circle collision model are as follows:

[0011] 3.1) Divide the unmanned vehicle projected onto the local grid map into four equal-sized medium rectangles along the length direction;

[0012] 3.2) Divide the middle rectangles at both ends of the unmanned vehicle into two small rectangles of the same size along the width direction of the unmanned vehicle;

[0013] 3.3) With the center point of the unmanned vehicle as the center, build the large circumscribed circle of the unmanned vehicle. With the center points of the two middle rectangles in the middle of the unmanned vehicle as the center, build the middle circumscribed circle of the middle rectangle. With the center points of the four small rectangles at both ends of the unmanned vehicle as the center, build the small circumscribed circles of the small rectangles to complete the seven-circle collision model.

[0014] 4) Collision check;

[0015] The discrete points of the path are used as the projection center points of the unmanned vehicle. If the distance to the nearest obstacle of the unmanned vehicle is greater than the radius of the large circumscribed circle, the unmanned vehicle will not collide. If the distance to the nearest obstacle of the unmanned vehicle is less than or equal to the radius of the large circumscribed circle, the distance to the nearest obstacle of the unmanned vehicle and the radii of all the middle circumscribed circles and small circumscribed circles are judged. If the distance to the nearest obstacle of the unmanned vehicle is greater than the radii of all the middle circumscribed circles and small circumscribed circles, the unmanned vehicle will not collide, otherwise the unmanned vehicle will collide.

[0016] Preferably, in step 1), the step of establishing a local grid map is:

[0017] 1.1) Collect point clouds through the unmanned vehicle’s laser radar and camera respectively;

[0018] 1.2) The point cloud collected by the lidar is fused with the point cloud collected by the camera through Kalman filtering to form a point cloud fusion map;

[0019] 1.3) Remove the ground road point cloud from the point cloud fusion image;

[0020] 1.4) The point cloud fusion image after removing the ground road point cloud is projected into the two-dimensional plane grid image of the vehicle body coordinate system, and the obstacles are marked in the grid image by the binarization method.

[0021] Preferably, in step 1.3), the ground road point cloud is removed by using a region growing method.

[0022] Preferably, in step 1.4), the process of marking obstacles by the binary method is as follows: if a grid in the grid map has a projected point cloud, the grid is an obstacle and is marked in black; if a grid in the grid map has no projected point cloud, the grid is not an obstacle and is marked in white.

[0023] Preferably, in step 2), an opencv distance transformation function is used to calculate the distance between each grid and its nearest obstacle grid.

[0024] The advantages of the present invention are: the seven-circle collision model of the present invention constitutes a two-layer collision detection standard, and the large circumscribed circle is used as the first layer of rough detection, which can quickly determine whether the unmanned vehicle collides with an obstacle, greatly reducing the amount of calculation and saving computing resources, while the remaining six circumscribed circles inside cover the space occupied by the unmanned vehicle to the greatest extent, and the six internal circumscribed circles are used as the second layer of fine detection, which further improves the accuracy of unmanned vehicle collision detection, increases the safety of unmanned vehicle driving, and has a small amount of calculation. The present invention can meet the needs of real-time collision detection required by unmanned vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of the present invention;

[0026] Figure 2 This is the seven-circle collision model of the present invention. DETAILED DESCRIPTION

[0027] See also Figure 1 , a collision detection method for an unmanned vehicle based on the seven-circle method, comprising the following steps:

[0028] 1) Establish a local grid map of the environment around the unmanned vehicle based on the laser radar and camera respectively;

[0029] 1.1) Collect point clouds through the unmanned vehicle’s laser radar and camera respectively;

[0030] 1.2) The point cloud collected by the lidar is fused with the point cloud collected by the camera through Kalman filtering to form a point cloud fusion map;

[0031] 1.3) Use the region growing method to remove the ground road point cloud in the point cloud fusion image;

[0032] 1.4) The point cloud fusion image after removing the ground road point cloud is projected into the grid image of the two-dimensional plane of the vehicle body coordinate system. The obstacles are marked in the grid image through the binarization method. If the grid in the grid image has a projected point cloud, the grid is an obstacle and is marked in black. If the grid in the grid image has no projected point cloud, the grid is not an obstacle and is marked in white. The binarization method can quickly mark the position of the obstacle, which is convenient for subsequent collision detection.

[0033] 2) Traverse all grids in the local grid map, use the opencv distance transformation function to calculate the distance between each grid and its nearest obstacle grid, and store the distance in the grid. The opencv distance transformation function is easy to call and does not require additional algorithm design, reducing the workload;

[0034] 3) Project the unmanned vehicle onto the local grid map, and establish a seven-circle collision model on the unmanned vehicle projected onto the local grid map, see Figure 2 , the steps of establishing the seven-circle collision model are:

[0035] 3.1) Divide the unmanned vehicle projected onto the local grid map into four equal-sized medium rectangles along the length direction;

[0036] 3.2) Divide the middle rectangles at both ends of the unmanned vehicle into two small rectangles of the same size along the width direction of the unmanned vehicle;

[0037] 3.3) With the center point of the unmanned vehicle as the center, build the large circumscribed circle of the unmanned vehicle. With the center points of the two middle rectangles in the middle of the unmanned vehicle as the center, build the middle circumscribed circle of the middle rectangle. With the center points of the four small rectangles at both ends of the unmanned vehicle as the center, build the small circumscribed circles of the small rectangles to complete the seven-circle collision model.

[0038] 4) Collision check;

[0039] The discrete points of the path are used as the projection center points of the unmanned vehicle. If the distance to the nearest obstacle of the unmanned vehicle is greater than the radius of the large circumscribed circle, the unmanned vehicle will not collide. If the distance to the nearest obstacle of the unmanned vehicle is less than or equal to the radius of the large circumscribed circle, the distance to the nearest obstacle of the unmanned vehicle and the radii of all the middle circumscribed circles and small circumscribed circles are judged. If the distance to the nearest obstacle of the unmanned vehicle is greater than the radii of all the middle circumscribed circles and small circumscribed circles, the unmanned vehicle will not collide, otherwise the unmanned vehicle will collide.

[0040] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification made to the present invention by those skilled in the art without departing from the spirit of the present invention shall fall within the protection scope of the present invention.

Claims

1. A collision detection method for an unmanned vehicle based on the seven-circle method, characterized in that: The following steps are involved: 1) Establish a local grid map of the environment around the unmanned vehicle based on the laser radar and camera respectively; 2) Traverse all grids in the local grid map, calculate the distance between each grid and its nearest obstacle grid, and store the distance in the grid; 3) Projecting the unmanned vehicle into the local grid map, and establishing a seven-circle collision model on the unmanned vehicle projected into the local grid map, wherein the steps for establishing the seven-circle collision model are as follows: 3.1) Divide the unmanned vehicle projected onto the local grid map into four equal-sized medium rectangles along the length direction; 3.2) Divide the middle rectangles at both ends of the unmanned vehicle into two small rectangles of the same size along the width direction of the unmanned vehicle; 3.3) With the center point of the unmanned vehicle as the center, build the large circumscribed circle of the unmanned vehicle. With the center points of the two middle rectangles in the middle of the unmanned vehicle as the center, build the middle circumscribed circle of the middle rectangle. With the center points of the four small rectangles at both ends of the unmanned vehicle as the center, build the small circumscribed circles of the small rectangles to complete the seven-circle collision model. 4) Collision check; The discrete points of the path are used as the projection center points of the unmanned vehicle. If the distance to the nearest obstacle of the unmanned vehicle is greater than the radius of the large circumscribed circle, the unmanned vehicle will not collide. If the distance to the nearest obstacle of the unmanned vehicle is less than or equal to the radius of the large circumscribed circle, the distance to the nearest obstacle of the unmanned vehicle and the radii of all the middle circumscribed circles and small circumscribed circles are judged. If the distance to the nearest obstacle of the unmanned vehicle is greater than the radii of all the middle circumscribed circles and small circumscribed circles, the unmanned vehicle will not collide, otherwise the unmanned vehicle will collide.

2. The unmanned vehicle collision detection method based on the seven-circle method according to claim 1, characterized in that: In the step 1), the steps of establishing a local grid map are: 1.1) Collect point clouds through the unmanned vehicle’s laser radar and camera respectively; 1.2) The point cloud collected by the lidar is fused with the point cloud collected by the camera through Kalman filtering to form a point cloud fusion map; 1.3) Remove the ground road point cloud from the point cloud fusion image; 1.4) The point cloud fusion image after removing the ground road point cloud is projected into the two-dimensional plane grid image of the vehicle body coordinate system, and the obstacles are marked in the grid image by the binarization method.

3. The unmanned vehicle collision detection method based on the seven-circle method according to claim 2, characterized in that: In the step 1.3), the regional growing method is used to remove the ground road point cloud.

4. The unmanned vehicle collision detection method based on the seven-circle method according to claim 2, characterized in that: In the step 1.4), the process of marking obstacles by the binary method is as follows: if a grid in the grid map has a projected point cloud, the grid is an obstacle and is marked in black; if a grid in the grid map has no projected point cloud, the grid is not an obstacle and is marked in white.

5. The unmanned vehicle collision detection method based on the seven-circle method according to claim 1, characterized in that: In the step 2), the distance between each grid and the nearest obstacle grid is calculated using the opencv distance transformation function.

Images