Measurement of the size distribution of rocks on a conveyor belt using machine vision

Abstract

This research is concerned with the online measurement of the size distribution of rocks on a conveyor belt. The measurement process developed In this thesis is based upon experimental work as a consequence of intractable mathematics developed when extrapolating models of Meal particle systems to rocks on the conveyor belt. Initial experimentation and background theory motivate the use of machine vision incorporating two-dimensional image processing and analysis techniques for the rocksize measurement The measurement process thus consists of images of rcckscenes which are processed to produce traces, from which populations of chord-length intercepts are measured and finally converted to size distributions. The traces or rock boundaries are produced by combining edge detection, thresholding and morphological techniques. Because of occlusion and the random nature of the rock patterns, the edge detection and segmentation process es error prone. Thus the boundary patterns must be subjected to edge correction to produce corrected segmented-edge patterns from which chord-lengths can be measured. The rockstream on the conveyor belt is a stochastic process where each image and associated chord-length distribution is a sample of the random variable rocksize distribution. Thus because of the stochastic nature of the rockstream and rock-segmentation errors, existing analytical and inversion techniques produce meaningless size distributions when transforming from chord-Iength, Thus a new approach is developed where the chord- length distributions are processed statistically with moment or morphological methods to transform the populations of chord-length distributions into meaningful size data. The first moment of the measured chord-length distribution allows for the determination of average-sieve size with a variance of between 5-25% over a size range of 6:1. Weighted moment methods produce distributions which are visually and qualitatively indicative of the sieve size distribution of rocks on the conveyor belt Weighted moment method also present data from which the proportion by mass of larger rocks to smaller rocks may be estimated