During process of objects tracking, problem of tracking box about marked objects is a major problem. Moreover, tracking of multi-objects are also difficult problems of objects tracking. This paper can tag object automatically through using motion vector of Mpeg2 to mark activities object of static video. Then, we extract multi-dimensional characteristics from initial goal of motion vector determined and made model. And accurately identify particles of larger weight to achieve purpose of accurately tracking objects through process of original value of particle filter matching observed value. As adopted methods of pattern classifying, so made feature matching between new particle and original particle are more accurate. The experiments show that the algorithm had good tracking performance and strong robustness.