To solve the problems of tracking failure caused by blocke d target and low tracking precision in the tracking learning detection (TLD) a lg orithm, an improved TLD algorithm based on multi innovation Kalman filter is pr o posed. The improved algorithm models the target before tracking, u ses the results of TLD tracking algorithm as the current observations and then c ombines with the predicted values of the multi innovation Kalman filter to opti m ize the tracking results. The experiments show that the improved algorithm of TL D has higher precision than the original TLD algorithm, and it is able to predict the position of the target when the target is occluded.