Multiscalar rock recognition using active vision

This paper describes an active vision approach to rock recognition for the purpose of measuring rock size distributions or fragmentation. Such a measure is useful in the mining industry to facilitate the control and monitoring of autogenous mills. A new approach for recognising and segmenting rock images is described. It is based on rapid identification of targets at which attention is focused for analysis. In addition, a multiscalar image pyramid is used in a novel way to allow target detection and analysis to be optimised for a particular size, thereby improving their performances. The results that have been obtained with this approach are both robust and accurate.