Abstract: In view of problems of poor real-time performance and incomplete coverage of traditional dust monitoring methods, two kinds of design scheme of dust monitoring system based on machine vision were proposed, namely dust monitoring systems based on monocular vision and binocular vision. The dust monitoring system based on monocular vision uses frame difference method and corrosion expansion algorithm to realize rapid recognition of the dust target in the field of view. Based on monocular vision, the dust monitoring system based on binocular vision uses calibration target and three-dimensional space reconstruction to achieve dust positioning. The experimental results show that the dust monitoring system based on monocular vision can capture formation process of dust cluster, and the real-time processing rate is four frames per second; the dust monitoring system based on binocular vision can further measure the position information of dust clusters, and positioning error is less than 10%.