To solve the problem that the trajectory based on traditional pedestrian tracking method is very different from the real trajectory, this paper proposed an improved particle filter pedestrian tracking method based on 3D model. It used camera calibration information and image sequence information to construct 3D pedestrian model, in order to deal with scale variation and extract real trajectory. And double exponential smoothing is used to improve particle filter, so it can deal with occlusion issue and reduce computation complexity. The experiment results of this paper indicates that the proposed method can cope with the situations of targets occlusion and scale variation. It has good performance in pedestrian tracking compared to standard particle filter as well as KPF.