Edge registration versus triangular mesh registration, a comparative study

Abstract

This paper presents a comparative study of two techniques for registering range images. The first one consists in registering range images represented by means of compact adaptive triangular meshes. The second approach registers the same range images but represented by means of their edges. In both approaches the ICP algorithm is used to compute the registration parameters (rotation and translation). The main objective in both approaches is to register a compact representation instead of all original data points. The proposed comparative study is performed in an experimental way by using a set of real range images considering both structured and sculptured objects. Four different criteria are taken into account to perform the comparison: (1) robustness with respect to initial conditions (estimated relative positions between the range images to be registered), (2) robustness with respect to resolutions of set of points on which the ICP method is executed, (3) robustness with respect to overlapping between view fields, and (4) number of iterations versus registration error. By employing these criteria, triangular mesh based and edge based registrations are tested on several range images and results are compared with respect to the ground truth. Conclusions from these experimental results are presented.