Depth gradient image based on silhouette -: A solution for reconstruction of scenes in 3D environments

Greatest difficulties arise in 3D environments when we have to deal with a scene with dissimilar objects without pose restrictions and where contacts and occlusions are allowed. This work tackles the problem of correspondence and alignment of surfaces in such a kind of scenes. The method presented in this paper is based on a new representation model called Depth Gradient Image Based on Silhouette (DGI-BS) which synthesizes object surface information (through depth) and object shape information (through contour). Recognition and pose problems are efficiently solved for all objects of the scene by using a simple matching algorithm in the DGI-BS space. As a result of this the scene can be virtually reconstructed. This work is part of a robot intelligent manipulation project. The method has been successfully tested in real experimentation environments using range sensors.