Hand Detection And Tracking For Human-Computer Interaction

With the rapid development of information technology,especially the machine learning and pattern recognition based image processing technology is becoming more and more mature,gesture control and recognition gradually replace the mouse,keyboard and become an important way of interaction in human-computer interaction.Hand detection and tracking is the technical basis for gesture control.In this paper,we study the hand detection and tracking algorithms,then proposed three related algorithms,which are hand detection based on multi clues,hand tracking using exemplar-based KCF tracker,and hand tracking combining superpixel segmentation.The hand detection algorithm based on multiple clues,which exploits motion cues,cascade classifiers and time clues to solve the effective and efficient problem in hand detection.The experiments result shows that the proposed algorithm has achieved good results in various complicated environemt.The hand tracking using exemplar-based KCF tracker,which combines optical flow and feature points to estimate candidate position in current frame.During the tracking process,multiple KCF trackers constitute a tracker ensemble and each tracker can generate a candidate position,so we can obtain a bag of candidates.Finally,the combination of the color model and the tracker ensemble is used to rate all the candidate positions,and the highest candidate position is determined as the final hand position.The hand tracking using exemplar-based KCF tracker,which can not obtain the accurate hand position,so the tracking result may deviate from the ground truth which may lead to model drift during update process.Aiming at the shortcomings of this algorithm,the hand tracking combining superpixel segmentation,which is based on the algorithm that hand tracking using exemplar-based KCF tracker,and superpixel segmentation is introduced to obtain the accurate hand position.In the proposed algorithm,the base tracker is used to obtain the hand position,and then the hand position and its surrounding area are traken as region of interest.In this region,the superpixel segmentation based on the color model is carried out to obtain the hand contour and determine the accurate position of the hand.