Research On Image Fusion Method Of Vehicle Point Cloud And Street View Image

Three-dimensional laser scanning technology has developed into a relatively mature new high-efficiency mapping technology,which can quickly and accurately acquire a large amount of point cloud data on the surface of the target without touching the target,with less interference from external environmental factors and uniform accuracy.However,due to its lack of texture and color information,it is not easy to interpret and analyze.The image can be used to extract the shape and color of the target object,and directly obtain the texture information of the object,which has strong complementarity with the laser scanning data.Therefore,the fusion research of laser point cloud data and image data has strong practical value.In this paper,the vehicle laser scanning point cloud and street scene image are the main data sources.The motion recovery structure theory,the scale conversion estimation idea and the four-point consistent basis registration strategy are used to study the vehicle point cloud and street scene image fusion methods.The main contents are as follows: :(1)Based on the basic principle of motion recovery structure,the three-dimensional reconstruction method of street scene image is studied.SIFT feature extraction is used to realize street scene image matching and generate sparse connection points,and then the SFM dense point cloud model corresponding to street scene image is obtained,which provides indirect registration data source for street scene image and vehicle laser scanning point cloud fusion.The SFM point cloud model based on principal component analysis and the scale conversion method of vehicle laser point cloud data are studied,which provides a solution for point cloud data registration with different scales.(2)The SFM point cloud model based on principal component analysis and the scale conversion method of vehicle laser point cloud data provide solutions for point cloud data registration with different scales.(3)From coarse to fine image and point cloud indirect registration strategy.The four-point uniform basis is used as the registration primitive to realize the rough registration of the in-vehicle laser point cloud and the SFM dense point cloud model,which provides a better initial position for further ICP fine global registration and avoids localization during the registration process.(4)Experiment verification and analysis of the on-board point cloud and street view image registration methods.