Registration of 3-D partial surface models using luminance and depth information

Textured surface models of three-dimensional objects are gaining importance in computer graphics applications. These models often have to be merged from several overlapping partial models which have to be registered (i.e. the relative transformation between the partial models has to be determined) prior to the merging process. In this paper a method is presented that makes use of both camera-based depth information (e.g. from stereo) and the luminance image. The luminance information is exploited to determine corresponding point sets on the partial surfaces using an optical flow approach. Quaternions are then employed to determine the transformation between the partial models which minimizes the sum of the 3-D Euclidian distances between the corresponding point sets. In order to find corresponding points on the partial surfaces luminance information is linearized. The procedure is iterated until convergence is reached. In contrast to only using depth information, employing luminance speeds up convergence and reduces remaining degrees of freedom (e.g. when registering sphere-like shapes).