Overspeed is a major factor causing road traffic accidents in China. Since the early 1980s, researchers have developed various algorithms to extract speed information from traffic image sequences. In this paper, an algorithm is proposed to evaluate the speed of the accident vehicle by videos. With human interaction, control points are refined by corner detector and then used for two-dimension geometric correction. The frame difference method is employed to capture the accident vehicle in images. Once the velocity curve is obtained and smoothed, the motion trajectory and speed of an accident vehicle at collision can be captured as well. Experimental results show that the proposed method is more accurate in speed measurement. It will facilitate the traffic police to deal with traffic collision cases in a real world. © 2010 IEEE.