Research On Road Edge Sensing Technology Of Outdoor Unmanned Sweeper

In the road edge sensing technology of mobile robots based on multi-line laser,constructing road edge features is a main way of road edge extraction at present.This topic combines the road condition of outdoor unmanned sweeper,uses neural network to learn road edge features,extracts road edge,and uses road edge points as feature points of Lidar SLAM to construct 3D Lidar SLAM system of outdoor unmanned sweeper.According to the research content of the subject,the research in this paper is mainly divided into the following parts:Road segmentation algorithm based on multi-plane iteration.This topic proposes a multi-plane iterative pavement segmentation algorithm to segment 3D Lidar point clouds.Due to the large amount of original point cloud data and the redundancy of many spatial point clouds,the focus of research is road edge perception,so road surface point clouds need to be separated.Because the actual road surface has obvious but not absolute plane geometric characteristics,a plane model estimation method with prior knowledge is proposed to fit the road surface and estimate the road surface point cloud.Aiming at the uneven problems such as local slopes,pits and the like existing in the actual road surface,a segmentation idea is proposed to estimate the plane model of each small segment point cloud and finally splice into a complete road surface.The algorithm can separate complex road surfaces including straight roads,curves,ramps and fork roads,and is more stable and faster than the traditional road segmentation algorithm.Road edge extraction network based on graph convolution.A graph convolution network is proposed for Road Curb Detection Net)RCDNet)to transform road curve point extraction into point cloud segmentation.Traditional artificial road edge features have great limitations,which can only deal with a certain fixed road condition and have low accuracy for road edge extraction under multiple road conditions.In order to solve the problem of road edge extraction,this paper proposes a neural network framework of edge graph convolution and automatic road edge feature extraction.By fusing local information and global information of point cloud in high-dimensional space,more neighborhood information of point cloud can be obtained,and each point can be classified,so that the segmentation accuracy of the network for three-dimensional point cloud is higher.The network framework can extract road edge points under various road conditions stably and has high extraction accuracy.The 3D Lidar SLAM system based on RCDNet.The extraction of general feature points is more stable and accurate under the condition of more environmental features.Combined with the actual working conditions of unmanned vehicles,due to the continuous and stable characteristics of road edge points,a method of using road edge points extracted by RCDNet network as feature points is proposed,which can more accurately construct laser odometer,and correct the estimation error through factor graph optimization method to realize the incremental map construction of outdoor unmanned vehicles.