Registration of surfaces minimizing error propagation for a one-shot multi-slit hand-held scanner

Abstract

We propose an algorithm for the on-line automatic registration of multiple 3D surfaces acquired in a sequence by a new hand-held laser scanner. The laser emitter is coupled with an optical lens that spreads the light forming 19 parallel slits that are projected to the scene and acquired with subpixel accuracy by a camera. Splines are used to interpolate the acquired profiles to increase the sample of points and Delaunay triangulation is used to obtain the normal vectors at every point. A point-to-plane pair-wise registration method is proposed to align the surfaces in pairs while they are acquired, conforming paths and eventually cycles that are minimized once detected. The algorithm is specially designed for on-line applications and can be classified as a closing-the-loop technique, where there are not that many competing methods, though it has been compared to the literature. Experiments providing qualitative and quantitative evaluation are shown by means of synthetic and real data and we demonstrated the reliability of our technique.