A novel algorithm is presented in this paper to detect anomalous vehicle behaviors such as abnormal stop and vehicle crashing. After objects detection, spatio-temporal trajectory of multiple objects can be obtained to construct the regional short-time constitute velocity model. Then gray model theory is adopted to estimate the motion model parameters. With the definition of abnormal traffic events patterns, individual vehicle movement can be judged as an abnormal or normal vehicle behavior. Experimental results show the proposed method can detect the abnormal traffic events effectively.