Research On The Method Of High-Definition Map Creation In Congested Environment

With the gradual development of automatic driving technology,higher requirements are put forward for the refinement,semantics and customization of map.The traditional navigation map can not meet the needs of automatic driving.As an electronic map with high precision and fine definition,high-definition map has become a necessary condition for driverless technology.The generation of high-definition map mainly includes visual image and laser point cloud.Because the image acquisition speed is fast,the acquisition space is small and the semantics is obvious,it is the main acquisition method of the ground traffic signs at present.However,when collecting maps,it is inevitable to encounter a congested environment.Social vehicles will block the camera vision around,resulting in the loss of ground information,affecting the collection effect.During the acquisition process,the image location is inaccurate due to the location jitter.At the same time,the traditional image overlay makes the collected information not fully utilized,which leads to the low definition of map.This paper proposes a high-definition map building method for congestion environment.The main research work of this paper includes:First of all,the demand of high-definition vector map for autopilot is systematically analyzed.Based on this,a whole set of high-definition map construction scheme is designed,including the definition of high-definition map,acquisition equipment and platform,data processing method and map generation and storage.Secondly,for the problem that the ground information is occluded by vehicles,based on the understanding of panoramic image,this paper uses the deep learning method to design a semantic segmentation scheme for panoramic image,which eliminates the information of vehicles and pedestrians on the road,and uses the spatial consistency to complete the occluded road surface,so as to recover the occluded information.Thirdly,for the problem of location error,this paper first removes the dynamic vehicle and shadow in the image,and then proposes a top view matching and fusion location method based on the road surface.Through the fusion of image matching,odometer and GPS positioning results,the location error of map layer is greatly reduced.Finally,aiming at the problem of layer blur caused by image superimposition,this paper proposes an image fusion method based on wavelet transform,which can improve the overall accuracy and clarity of the ground orthophoto image.Experiments show that this method is effective.Then,this paper proposes a vector map construction scheme based on this,which shows the existing work.In general,this paper designs a high-definition map making system for traffic congestion environment,and studies and solves the problems of vehicle occlusion,positioning error and layer blur in the process of map building.