Abstract: Classical Mean-Shift algorithm requires that the adjacent two frames target template must have overlap areas.In order to resolve this weak point，an improved algorithm combined with Kalman filter is proposed.First，Kalman filter predicts a position which is to be the initial search center in Mean-Shift algorithm to track.Then，Mean-Shift gets a new target position which is used to be the input parameter of the next Kalman filter.Experimental results show that the algorithm not only can resolve problem which the adjacent two frames target template areas do not overlap because of target velocity changing suddenly or target moving fastly，but also has good performance in ordinary shelter problem.