Range and Intensity Vision for Rock-Scene Segmentation

This paper presents a methodology for the automatic segmentation of rock-scenes sing a combination of range and intensity vision. A major problem in rock scene segmentation is the effect of noise in the form of surface texture and color density variations, which causes spurious segmentations. We show that these problems can be avoided through pre-attentive range image segmentation followed by focused attention to edges. The segmentation process is inspired by the Human Visual System's operation of using a priori knowledge from pre-attentive vision for focused attention detail. The result is good rock detection and boundary accuracy that can be attributed to independence of range data to texture and color density variations, and knowledge driven intensity edge detection respectively. Preliminary results on a limited image dataset are promising.