Research On The Detection Technology Of Obstacles In Non-structural Environment For Construction Vehicles

The environment perception technology is the foundation for achieving unmanned driving and autonomous operation of construction vehicles in unstructured environments.Due to the unclear boundary between the unstructured environment and the terrain,and there are many irregular objects attached to the ground,it is difficult to describe the environment information in the structured and regular language,which makes it difficult to perceive the environment,especially detect obstacles.As a widely used sensor in the intelligent driving environment sensing system,the lidar can directly obtain the three-dimensional information of the environment.Compared with the camera,the lidar is not easily affected by light.Compared with millimeter wave radar,lidar is more accurate in ranging.This article combines the project of the National Natural Science Foundation of China "Intelligent loading and unloading strategy and multi-objective trajectory planning for unmanned mining shovels" to conduct research on the relevant technology of obstacle detection using lidar in unstructured environments for construction vehicles.An obstacle detection algorithm suitable for unstructured environments is proposed,and the principle of the proposed method is verified through physical prototype experiments of loaders.Firstly,the research status of point cloud filtering and segmentation and obstacle detection algorithms at home and abroad are introduced.In view of the point cloud distortion caused by vibration and other problems during the driving operation of construction vehicles,the causes and effects of point cloud distortion are analyzed in detail,and a method for correcting point cloud distortion is proposed by applying integrated navigation to provide corresponding data to unify all points in a frame of point cloud data into the same coordinate system.The filtering effect of traditional filtering algorithm in unstructured environment is analyzed,and an improved filtering algorithm is proposed according to the characteristics of unstructured environment.This paper analyzes the principle of common ground segmentation algorithms,and puts forward a ground point cloud segmentation algorithm based on region division in unstructured environment,which comprehensively considers the multiple characteristics of local point clouds and combines the trafficability of construction vehicles to achieve ground point cloud segmentation.The proposed ground point cloud segmentation algorithm can achieve good segmentation results in unstructured environments.Apply the connected region labeling algorithm applied in the field of image processing to point cloud data processing for non-ground point cloud clustering in unstructured environments.After analyzing the point cloud data of two consecutive frames,it is concluded that the dynamic obstacle detection can be achieved by measuring the spatial similarity of the two point cloud clusters clustered in the point cloud data of two consecutive frames corresponding to the same obstacle after coordinate transformation.The proposed multi-feature-based point cloud association algorithm is applied to associate the point cloud clusters corresponding to the same obstacle in two consecutive frames of point cloud data,and then the double-buffered octree algorithm is applied to detect dynamic obstacles based on spatial similarity.Finally,the proposed dynamic obstacle tracking algorithm based on recursive confidence judgment is applied to further optimize the detection performance of dynamic obstacles.The proposed obstacle detection method is tested on a physical prototype of loader.Relevant algorithms are verified in the experimental site of unstructured environment,and the results showed that the proposed algorithm can reliably detect obstacles on the basis of ensuring real-time performance.The research work in this article has practical significance for achieving unmanned driving of construction vehicles.