An Environment Perception Method Based On The Fusion Of Stereo And Lidar

With the development of automatic driving technology,the automatic driving vehicle has been applied in many walks of life,such as road cleaning,short-distance transportation and passenger transportation.At the same time,in order to deal with the complex environment faced by different tasks,the safety of automatic driving is also put forward higher requirements.In the component unit of automatic driving technology,environmental perception is an important part,which is the guarantee of its security and intelligence.Lidar and stereo vision are commonly used in environmental perception.Lidar has the advantages of high accuracy and strong stability,and can obtain accurate three-dimensional information in complex environment.However,due to the sparsity of laser point cloud,some environmental information is often lost.At this time,the advantages of the density of stereo vision are reflected.Compared with lidar,stereo vision has the advantages of low cost,high resolution and can obtain dense disparity map.Considering the respective advantages of stereo vision and lidar,this paper mainly designs a road environment perception method based on stereo vision and lidar,and obtains accurate and dense disparity map by fusing the information collected by the two methods.At the same time,because autopilot needs to face a variety of complex scenes,how to design a robust auto exposure algorithm to collect information rich images is also one of the research focuses.My main work is as follows:In stereo vision,the quality of the image will have a great impact on the accuracy of the followup algorithm.Therefore,we need to ensure that in the complex environment,through effective exposure time adjustment,we can obtain the image with rich details and clear texture.In this paper,a automatic exposure algorithm based on feature detection is designed,which guarantees the image quality of texture rich areas and can adapt to the changes of environment in different scenes.Compared with the camera’s own auto-exposure algorithm,the method in this paper has better performance in high dynamic environment.After obtaining high quality images,dense disparity information is obtained by stereo matching algorithm,and then fusion is carried out with lidar data.Because of the inaccuracy of stereo vision and the sparsity of lidar,an end-to-end convolution neural network is designed in this paper.By inputting stereo disparity map and lidar disparity map,the two are fused to obtain more dense and accurate disparity map,improve the accuracy of environmental perception,and provide an accurate depth estimation for the whole system.