We present a box depalletizing system based on images acquired with a time of flight laser sensor mounted on the hand of the robot. Scanning the upper layer of the pallet yields a 2.5D image to which edge detection and robust line fitting are applied to extract 3D vertices. This vertex information is used as an input for a model based object recognition system. Model vertices are matched to scene vertices and object location hypothesis are formed. These hypotheses are verified or rejected using a two-step verification process. Due to the fact that we use edges to extract vertices, rather than surfaces, we are able to detect target objects, in both cluttered and ordered configurations. Our experiments with different configurations of cardboard boxes and paper tissue packets demonstrate the validity of our approach. The main advantages of our system are it's versatility, simplicity, efficiency and robustness.

Object oriented representation of image sequences requires accurate motion segmentation and depth ordering techniques. Unfortunately, the lack of precise motion estimates at the object boundaries makes these two tasks very difficult. We present a detailed analysis of the behaviour of dense motion estimation techniques at object boundaries which reveals the systematic nature of the motion estimation error; the motion of the occluding surface is observed in a small neighbourhood on the occluded side. We then show how the joint use of still image segmentation and robust regression can eliminate this error. Furthermore we present a novel technique which uses the position of the error as a depth cue. The validity of this technique, which requires only sub-pixel motion and which is capable of distinguishing between different types of intensity discontinuities, such as object boundaries, surface marks and illumination discontinuities, is then demonstrated on several synthetic and real image sequences.