Research On Visual Tracking Algorithm For The Applications Of Industrial Automatic Assembly Line

Vision tracking in industrial applications has made a lot of achievements, mainlyconcentrated in the automobile and engineering machinery, electronic, welding, etc. Inthis paper, moving target on the conveyor belt in industrial application is selected as themain research object, then designs a visual tracking system: First, detect the object byusing background-difference detection algorithm based on Gaussian mixture model; then,identify the target by using recognition algorithm based on contour feature; finally trackthe target by using the mean shift tracking algorithm combined with Kalman filtering andthe tracking algorithm based on improved Camshift algorithm. The mainly contents ofthis paper are as follows:(1) Three common target detection algorithms in existing industrial assembly lineare researched and compared. In this paper, the background-difference detectionalgorithm based on Gaussian mixture model is proposed. The experiments result showthat the proposed algorithm can detect moving targets better than others in industrialproduction line whose background is changed.(2) Because it is difficult to describe target features in the plane visual image usingthe geometry constants of perimeter, area, angle and diameter etc. In this paper, arecognition algorithm based on contour feature is presented, which can be used with anyshape of target. Experimental results indicated that the presented algorithm has goodcharacteristics of accuracy and real-time.(3) The mean shift tracking algorithm target is easy to track failure when themoving targets is completely covered. To solve this problem, a mean shift trackingalgorithm combined with Kalman filtering is proposed. The algorithm can judge andprocess the occlusion problem. This algorithm use the residual error between theobservations of the current frame of mean shift algorithm and the best estimated value tothe target location of Kalman filter to determine the extent of block. The Camshiftalgorithm relied only on the color information of target to track object, which wouldvulnerably lost the target in the interference of the same color. To solve this problem, thealgorithm process interference of the same color mixed by Gaussian backgroundmodeling. In order to improve the algorithm anti-block ability, use the Kalman filter topredict the location of the next frame of the moving target. Also use the Kalman filter tofix the position of moving target. Experiments show that the mean shift tracking algorithm combined with Kalmanfiltering and the tracking algorithm based on improved Camshift algorithm caneffectively solve the occlusion and the same color interference problems, which meetsthe requirements of robustness and accuracy of target tracking on the assembly line in thegeneral industrial environment.