Drivable Area Detection Of Unstructured Roads Based On 3D Lidar

Drivable area detection refers to intelligent vehicle detection and extraction of the area on the lane where the driving route can be planned.It is an important research point in the field of autonomous driving perception.Unstructured roads pose great challenges to drivable area detection due to factors such as complex road conditions and unclear road edges.Currently,drivable area detection methods for unstructured roads mainly rely on cameras or 3D Li DAR sensors.Cameras are easily affected by weather and lighting factors,and their reliability is poor in low-light environments,so this type of method has great limitations.The detection method based on 3D lidar has the problems that it is difficult to correctly separate the ground point cloud when the road surface is undulating,and it cannot deal with unstructured road scenes where the boundary space features are not obvious,which makes this type of method unable to meet the driving standards of automatic driving.To this end,this paper proposes a 3D lidarbased detection method for unstructured road drivable areas.The specific work and innovations are as follows:(1)Aiming at the problem that it is difficult to correctly separate the ground point cloud in the unstructured road scene with undulating ground or overhanging obstacles,a ground point cloud extraction algorithm based on adaptive sector grid image is proposed.Firstly,the sector grid map is adaptively generated according to the angular resolution of the 3D lidar to solve the problem of uneven distribution of point cloud density in each grid of the traditional grid map;then,based on the point cloud elevation features in adjacent grids,the extraction process is divided into two steps: rough extraction of ground grids and fine extraction of ground point clouds,so as to improve the robustness of the algorithm.A comparative experiment was carried out on the KITTI public dataset and the self-recorded actual road dataset.The results show that the proposed algorithm has an average extraction accuracy increase of 3.43% compared with the traditional raster image-based extraction algorithm,and the processing time of each frame is about 3ms faster,which is better than existing methods.(2)Aiming at the problem that the unstructured road cannot correctly extract the drivable area due to the inconspicuous boundary space features,a drivable area detection method for unstructured roads based on the fusion of space and reflection intensity is proposed.First,the cylindrical coordinate system detection model based on spatial features is improved by fusing reflection intensity factors;then,the ring detection model is improved by using intensity and dimensionality reduction space detection and used in conjunction with the cylindrical coordinate system detection model,which greatly improves the detection accuracy of candidate points;finally,using a Bezier curve to fit the drivable area boundary.A comparative experiment was carried out on the KITTI public dataset and the self-recorded actual road dataset.The results show that the proposed method improves the detection accuracy by an average of 8.41%on unstructured roads and by 2.32% on structured roads compared with common methods.,and has good real-time performance,which is better than existing methods.(3)Based on the two proposed methods,a single-vehicle unstructured road drivable area detection system is designed and developed.Running the test on the smart car,the results show that the system can complete the detection under the premise of an average of 32 frames per second,and the detection success rate can reach more than 95%,which has good effectiveness and stability.The method in this paper constructs an adaptive sector grid map and fuses spatial and intensity features for detection,which provides a new solution for drivable area detection and has certain theoretical value.At the same time,the developed detection system has significantly improved the accuracy of the detection of unstructured road drivable areas,and has a good detection effect on structured roads,which solves practical problems in the field of automatic driving and has high application value.