Research On Structured Road Environment Perception Method Of Intelligent Vehicle Based On Lidar And Image

With the improvement of living standards,the number of cars in our country is growing rapidly.At the same time,traffic accidents and traffic jams are becoming more and more serious.As an effective way to solve these problems,smart cars are gradually coming into our vision.Environmental perception is the first step of driverless car which is developing in full swing.The correct perception is related to the success or failure of driverless car,and it is indispensable.In this paper,environment perception is based on lidar and image.The main research contents are as follows:1.Recognize the road boundary based on 3D point cloud and cluster the obstacles on the road.Proposes a median filter method for ordered polar point clouds,which can effectively reduce the noise of point clouds.It combines height difference feature method,constraint of polar radius and angle method,constraint of point clouds tangent method to get grid map.Proposes a method to identify the road boundary in grid map,which consists three steps: road boundary search,filter and seed boundary growth.DBSCAN density clustering method is used to cluster obstacles.2.Recognition of lane lines in structured road based on camera image.Firstly,the image is filtered,distorted and perspective transformed.After that the top hat transform is used to extract lane lines seed points.finally,classify lane lines seed points and fit it.In this chapter,a lane line search and classification method is proposed,which can effectively distinguish the required lane lines.In the part of lines tracking,a method of using historical results to build mask is proposed,which can greatly reduce the calculation amount of classification.3.Calibrate vehicle coordinate system,lidar coordinate system and image coordinate system,and then unify recognition results of each sensor in the vehicle coordinate system.The influence of vehicle roll and pitch on image perspective transformation is considered.Set constraints to judge the correctness of the recognition results.If the recognition results are correct,according to these information infer the location of the missing lane lines,and search again.4.Build a test platform to verify the research methods proposed in this paper.The experiments recognize road boundary and obstacle based on lidar,recognize lane lines based on image and fuse recognition results are carried out,which can accurately identify road boundary,lane lines,the location and size of obstacles.The platform not only completes the verification,but also that it takes up less computer resources and has high efficiency.