Reconstruction of complex environments by robust pre-aligned ICP

This paper proposes a technique for three-dimensional reconstruction of an underwater environment from range views acquired by an acoustic camera. The final target of the work lies in improving the understanding of a human operator driving an underwater remotely operated vehicle floating close to an offshore structure. Due to the narrow field of view and the absence of control of the sensor position, no information is available about the degree of over-lapping between the range images; further, speckle noise and low resolution make more difficult the registration process. In this paper, we propose a preprocessing method which gives a coarse alignment of range images prior to running the Iterative Closest Point (ICP) algorithm for the accurate registration of views pairs. The pre-alignment is based on the matching between the three-dimensional skeletons extracted from the images. A comparative analysis is presented where our method is compared with plain ICP, and with a technique based on principal components analysis