Research On Road Detection And Estimation Based On Lidar Point Cloud

The accurate and effective road extraction is the key prerequisite for decision-making and path planning of intelligent vehicles.So,it is a hot research field in intelligent vehicle environment perception technology.The method of road extraction based on image segmentation is nearly mature at present.However,it is not accurate enough in distance measurement.In recent years,lidar has been gradually favored by researchers because of its advantages of high ranging accuracy and fast response.In this thesis,lidar is used as the main sensor for road area extraction.The research of road detection and estimation algorithm in this thesis mainly includes the following aspects.1.The background and research status in the field of road extraction are analysised.And the existing problems are concluded in this thesis.Based on these analyses,an iterative algorithm framework which include road detection and road estimation to extract the road area is proposed.The road detection part extracts the points belongs to road area from raw point cloud.The road estimation part according to the result of road detection part to estimate the range and slope of road.This framework can provide a grid map marked with road area and road slope information for intelligent cars,which is a powerful support for the safe and stable driving of intelligent cars.2.In terms of road detection,according to the road morphology features and lidar imaging principle,a certain number of features are selected to extract points belonging to the road area in the point cloud.On this basis,an algorithm is proposed to cluster the separated road points into clusters,and then to screen the point cloud clusters by the result of road estimation,so as to further improve the stability and accuracy of point cloud segmentation.3.In the aspect of road estimation,grid map is used to represent the range and the slope information of road area.Measurements are firstly generated according to the result of road detection.Then,for each grid of grid map,a binary bayesian filter is used to determine whether it is belongs to the road area on the one hand,on the other hand,an alpha filter is used to estimate the elevation of this grid.4.According to the algorithm proposed in this thesis,the intelligent vehicle hardware platform was built,the road detection and estimation system was designed and implemented,and the vehicle experiment was carried out.The experimental data show that the system in this thesis has sufficient accuracy and real-time performance,and can meet the actual use requirements.