Recognition of Free-Form Objects in Complex Scenes Using DGI-BS Models

Object recognition in 3D scenes with occlusion means identifying an incomplete and unknown object, which is arbitrarily posed, in an object database. This hard computer vision problem is solved in this paper through a new 3D shape representation called depth gradient image based on silhouette (DGI-BS). DGI-BS can be used to obtain a complete model as well as a partial model of an object. This property allows us to use it in complex scenes where incomplete surfaces of objects are available. The complete DGI-BS version synthesizes surface information (through depth image) and shape information (through contour) of the whole object in a single image smaller than 1 mega pixel. Object recognition is carried out by means of a simple matching algorithm in the DGI-BS space which yields a correspondence point-to-point between scene and model. This method has been successfully tested in real scenes with no special restrictions using range sensors.