Robust object recognition based on regular framing and depth aspect image

摘要

A method for model-based object recognition of a cluttered range scene including multiple objects is proposed. Our method uses 'depth aspect image' for three-dimensional model representation. A depth aspect image is defined as an orientation standardized appearance from the original depth data of the object, which is transformed by the congruent transformation drawn by each hypothetical basis pair of three barycenter points of voxels. The database consists of model range data and these depth aspect images which are made by all possible pair of barycenter points. Object recognition and position estimation are achieved through two-dimensional image matching between the depth aspect image from the scene and the ones from the database. This image matching is based on the 'least quantile of residuals' which is robust against occlusion occurred in cluttered scene. Sparse barycenter points make the ICP algorithm and the verification process faster. The paper includes a formalization of the proposed method and some experimental results with real objects.