Real-Time Tracking of Non-Rigid Objects using Mean Shift(meanshift算法下非刚性物体跟踪).pdf

RealTime Tracking of NonRigid Ob jects using Mean Shift Dorin Comaniciu Visvanathan Ramesh Peter Meer Imaging Visualization Department Electrical Computer Engineering Department Siemens Corp orate Research Rutgers University College Road East Princeton NJ Brett Road Piscataway NJ Abstract Mean Shift Analysis A new method for realtime tracking of nonrigid ob We dene next the sample mean shift intro duce the je cts seen fr om a moving camera is pr oposed The cen iterative mean shift pro cedure and present a new the tral computational module is based on the mean shift orem showing the convergence for kernels with convex iterations and nds the most pr obable target p osition in and monotonic proles For applications of the mean the current fr ame The dissimilarity between the target shift prop erty in low level vision ltering segmenta model its color distribution and the target candidates tion see is expressed by a metric derived fr om the Bhattacharyya coecient The theoretical analysis of the approach Sample Mean Shift shows that it relates to the Bayesian fr amework while Given a set fx i g of n p oints in the d i=1 n pr oviding a pr actical fast and ecient solution The dimensional space R d the multivariate kernel density capability of the tracker to hand le in realtime p artial estimate with kernel K x and window radius band occlusions signicant clutter and target scale varia width h computed in the p oint x is given by tions is demonstrated for several image sequences n XK x x i