Abstract: A segmentation algorithm for digital image sequence based on spatio temporal continuous wavelet transform (CWT) is developed. A separable motion filter is used to extract the motion feature parameters, and the segmentation based on the estimating of motion is achieved. Because of the separability of the wavelet function or motion filter, the computational complexity of multi dimentional CWT is decreased. Sequential searching and hierarchical clustering are used to improve the efficiency of segmentation. Different algorithms are used to do simulation by using real traffic image sequence, the simulation results illustrate that the segmentation algorithm based on motion feature is accurate and robust, even in the circumstance of lower signal noise ratio.