SEMI-AUTOMATIC REGISTRATION BETWEEN VEHICLE-BASED IMAGE SEQUENCE AND 2D VECTOR MAP

Since the street is a primary component of city, it is very significant for 3D city modeling (3DCM) that how to rapidly realize the 3D visualization of street sight. The acquisition of vehicle-based image sequence for buildings has the characteristics of convenience and rapidness. Moreover, the resolution of textures obtained from vehicle-based sequence is very high. Therefore, it will be aided greatly by the automatic processing of textures from vehicle-based image sequence that the 3D visualization of street sight in 3DCM. To reconstruct 3D texture model from images, the mapping from image space to object space should be acquired. Therefore, in this paper, image sequence and 2D vector map are registered for reconstruction of 3D texture model for urban street. Firstly, the image sequence is subsected automatically by the histogram of projective difference of corresponding points. Rectification and mosaic process are implemented to generate facade texture. After that, two corresponding point pairs between facade texture and map are selected to calculate the initial transform parameters from image coordinate system to map coordinate system. By those parameters, other corner points on the map are projected into image sequence. The positions of projected corner points, however, are only close to the exact corresponding points in raw images. Accordingly, the problem is how to extract those corresponding points. It is difficult to extract corner points directly from images because corner points may not be distinct features in close-ranged images and occlusio