Automatic Registration of Terrestrial Point Cloud Using Panoramic Reflectance Images

Much attention is paid to registration of terrestrial point clouds nowadays. Research is carried out towards improved efficiency and automation of the registration process. This paper reports a new approach for point clouds registration utilizing reflectance panoramic images. The approach follows a two-step procedure: pair-wise registration and global registrations. The pair-wise registration consists of image matching (pixel-to-pixel registration) and point clouds registration (point-to-point correspondence), provided the correspondence between image and point cloud (pixel-to-point) is known. The image matching process delivers corresponding points within the overlapping area of two images. These points are matched to their 3D equivalent points within the point clouds. False accepted correspondences are successfully removed by a geometric invariance check. An iterative least-square adjustment completes the pair wise registration. The global registration on all point clouds is obtained by a bundle adjustment using circularly self-closure constraint. The approach is tested with several data sets (indoor and outdoor scenes) acquired by the laser scanner FARO LS 880.