Research On Hand Detection And Motion Tracking Algorithm

With the development of intelligent human-computer interaction, hands as the naturalparts of human, human-computer interaction based on hands becomes an importantapplication, hands detection and motion tracking as the foundation of realization, in recentyears, make lots of progress but also some disadvantages. Based on videos or imagesequences, because of external factors, detection and tracking algorithm defects, it’s easy tolead to dynamic hand tracking lost or lag phenomenon, so how accurate real-time adaptivedetecting and tracking hands becomes the key to solve the problem.In hands motion tracking, based on HSV color histogram, using improved particle filteralgorithm to track hand automatically. Point to degradation and depleted phenomenon ofparticles in particle filter resampling phase, we propose the improved particle filteralgorithm based on maximum variance weights divided. Traditional particle filter trackingalgorithm, in order to determine the target center of the random scattering particles in theinitial frame, usually use manual calibration. In order to determine the initial position of thehands adaptively, using skin color clustering and Hu moments method combined for handsdetection and calibration automatically, finally in the initial frame circle hands with ellipsesand calculate the centers for hands tracking.Experiments show that, in hands detection method, color clustering and Hu momentscombined has a good effect on detection and extract of deformed hands, and is successfulfor hands tracking initialization, combined with improved particle filter algorithm based onHSV color histogram make the hands movement tracking be self-adaptive, and have acertain increase in the accuracy, stability and real-time.