Fusion-based color and depth image segmentation method for rocks on conveyor belt

Abstract

Semi-autogenous mill (SAG) feeding process requires accurate and reliable monitoring and controlling techniques for multi-sized rock distribution to stabilize crushing productivity. This study presents a color and depth (RGB-D) image fusion algorithm for the segmentation of multiscale rock images from the conveyor belt. The algorithm combines the multiscale characteristics of the color image and depth image, considers the extraction of large particles from the depth image and the rich texture details for segmentation of the color image, divides the segmentation regions according to different particle sizes, and finally obtains the contour of each particle after region-based superpixels merging. The proposed method outperforms any single sensor image segmentation method proposed in direct comparison, thereby attaining the closest to ground truth performance. The comparison results also show that the accuracy and adaptability of the algorithm are significantly improved, and this algorithm can therefore be used in rock-like material segmentation with multiple size distribution on conveyors.