Abstract: In view of problems that existing detection methods of operating state of vibrating screen can only detect local operating state of vibrating screen, and has shortcomings such as low precision and poor timeliness, an on-line detection method of whole operating state of vibrating screen based on binocular vision technology was put forward. Firstly, the method uses binocular vision detection device to capture movement images of the vibrating screen. Then it makes graying processing for the images， uses multi-scale Harris corner detection algorithm to obtain the feature points of the images, and introducing image pyramid theory to improve Lucas-Kanade optical flow estimation algorithm, so as to enhance on-line tracking ability of motion track for feature points. Finally, BP neural network classifier is designed to complete analysis and identification of motion trajectory of the feature points, so as to realize detection of whole running condition of vibrating screen. Test results show that the detection method has high accuracy and good timeliness, which can realize on-line detection and analysis of whole running state of vibrating screen with omni-directional and multi-angle tracking and identification for the trajectory of vibrating screen. Accuracy of the method respectively reached 97.917%, 90.667%, 96.431% and 93.181% when vibrating screen was in four kinds of states of stopping, normality and suspected fault and fault.