Clustering Method of 3D Point Cloud of Muck-Pile Based on Connectivity of Adjacent Surface

Abstract

This paper proposes a method to measure the fragmentation distribution of a pile of rocks (muck-pile) using image-based 3D reconstruction. One of the most important aspects of mine-blasting is appropriate rock fragmentation to optimize the cost of the blasting operation. The conventional method of measuring fragmentation distribution is based on 2D image processing including segmentation of muck-pile regions into rock clusters. However, in the 2D method, the measurement accuracy is limited. To accurately measure rock fragmentation distribution, we reconstructed a 3D model of a muck-pile from multi-view images and segment the 3D model based on rock-features such as color, normal vector, distance and adjacent angles of surface planes. As a result, the size of each rock was calculated by fitting a bounding box. Based on experimental evaluations, it was confirmed that the accuracy of the proposed method is higher than that of previous methods.