Geometric Correction And Stitching Method Of Multiple Street-view Panorama

In the information age with the continuous development of digital city,detailed 3D architectural models play an increasingly important role in 3D reconstruction of urban scenes.And realistic building facade texture is one of the important factors in 3D reconstruction.However,the facade images obtained by the existing methods are often insufficient in the aspect of fineness,which is usually difficult to meet the needs of detailed 3D reconstruction of digital cities.The mobile measurement system equipped with differential satellite positioning system,inertial measurement unit and street view panoramic camera can efficiently obtain high-resolution street panoramas of urban environment with accurate pose information,providing an important data source for 3d fine modeling of building facade structure.However,due to the unique projection mode of street view images,it is difficult to apply it directly to facade texture mapping.At the same time,due to factors such as ground occlusion,visual field limitation,image distortion,etc.,it is difficult for street view images under a single view to effectively cover the facade structure of large buildings.To solve the above problems,this paper proposes a geometric correction and stitching method for multiview street view images to eliminate the geometric and exposure inconsistencies among multiview images.The main research contents of this paper are listed as follow:1.Geometric correction of street view images based on gnomonic projection.The original streetscape image with equirectangular projection is transformed by gnomonic projection to restore the linear features of buildings in the image;2.Geometric registration of multi-view panoramic images based on SURF operator.The SURF operator was used to extract the feature points between the multi-view images,and the homography matrix was estimated after matching and purification of the feature points to complete the geometric registration of the multi-view images;3.Seamless stitching of multi-view panoramic images.After eliminating the exposure differences by using the gain compensation method,the multi-view images were searched for the optimal seamline,and multi-band blending was performed according to the search results to eliminate stitching lines between images and achieve seamless stitching.In order to verify the feasibility and effectiveness of the method in this paper,the open source street view image data in Google Maps is used as an example to conduct experiments.The results show that this method can solve the problem of the distortion of street view data sources and the limited coverage of single-view images.The stitching and optimized detailed images can be used as the facade texture data source of the 3D building model for mapping.And the texture accuracy of the facade reaches decimeter level,which provides important texture auxiliary data for the construction of the digital city.