In the light of the occlusion in the tracking of multiple vehicles, the article puts forward a new multi-object tracking algorithm based on feature correlation matching. The moving area that was extracted in current frame and in previous frame is respectively given different identity “vehicle region” and “temporary region”. The position of vehicles region in current frame can be forecast by using three points extrapolation, the distance between which and temporary region can be calculated and used to match current motion regions and track vehicles. In the meantime, the paper estimates states of vehicles depending on the state factors, and combines them with tracking windows based on segmentation method to solve vehicle tracking problem under occlusion efficiently. Experimental results indicate that the proposed algorithm can trace and forecast the vehicles which hided, and can accurately track object effectively.