Range and intensity vision for rock-scene segmentation

Abstract

This paper presents another approach to segmenting a scene of rocks on a conveyor belt for the purposes of measuring rock size. Rock size estimation instruments are used to monitor, optimize and control milling and crushing in the mining industry. We propose a joint analysis of range and intensity images of rocks where sections extracted from range images are used to guide a Model-driven intensity image segmentation process. The models contain rock shape and edge probability information which can be used to extract intensity edges of relevance in terms of orientation and position. Conventional polynomial fitting is used to extract the underlying rock shape as a final segmentation. Preliminary results on a small laboratory data set using qualitative and quantitative measures of performance are promising