Abstract
This paper describes a method of object recognition in terrestrial laser scan data of complex scenes. By local 3D shape matching between the CAD model mesh of the object and the laser scan data of the scene, the existence of the object is recognized and its location and orientation in the scene are extracted. Spin-images are used for shape matching. In this paper, some techniques for applying spin-image matching to the terrestrial laser scan data are proposed. They include robustness improvements for the scan noise and the differences in vertex densities by normal averaging and uniform point sampling, and efficient calculation by using multi-resolution images and geometric point filtering.