Research On Object Tracking Algorithm Based On Meanshift And Particle Filter

Target tracking Technique is one of key research in computer vision field, Itwas widely used in civil and military, including television monitors, bio-medicine,human-computer interaction, intelligent transportation, weapons guidance and so on.In recent years, domestic and foreign experts and scholars have put forward anumber of classical algorithms in order to track targets accurately and real-timely.Accurate and real-time tracking is still a challenge because of complex backgroundand high precise requirement.This paper studies MeanShift algorithm, Kalman filterand particle filter algorithm and proposes improved algorithms suitable for practicalapplication.The paper Studies the Meanshift algorithm which is a no parameter estimationmethod based on density gradient. It has Small amount of calculation and littleiterative times, but traditional Meanshift algorithm cannot guarantee trackingaccuracy in certain interference or occlusion case and background pixel in objectmodel will induce localization error. Aiming at the problem above a improved targetstracking algorithm based on Meanshift algorithm was proposed. The paper improvedtarget model and candidate model.Firstly, remove little probability characteristicvalue(such characteristic generally belong to non-target); Secondly, improve theobject model by introducing weights which decided by divisional degree betweenobject and background,and then the weight was used in object model to reduce the localization error of object tracking. Experimental results show that the new methodcan effectively and accurately track moving target in the cluttering background.The paper introduces the basic principles of correlation tracking and studieddeeply on classic Kalman filter algorithm. Then, the paper expounds in detail thecorrelation tracking algorithm based on Kalman filter prediction and evaluates theperformance of the algorithm according to experiments.The paper Studies the particle filter algorithm and aim at the problem that theParticle filter tracking algorithm costs huge computation and Kalman filter trackingalgorithm is no longer accurate in non-Gauss and non-linear case, a improved targetstracking algorithm based on Kalman filter and Particle filter was proposed. Firstly, acandidate object was gotten by Kalman tracking algorithm.Then, the tracking resultwould be verified by Particle filter algorithm when the match threshold is lower thana certain. The improved algorithm used "template buffer" to updated object templateto ensure the tracking process continuity, stability and accuracy.Experimental resultsshow that this approach can maintain the efficiency of Kalman algorithm and thepowerful ability of Particle filter algorithm, so it is of advanced property.