In order to improve the efficiency of obstacle avoidance of unmanned surface vehicles(USV), this paper presents a avoidance algorithm for unmanned surface vehicles USV to navigate safely in dynamic, cluttered environments. In the existing path planning algorithms, most of them take avoidance measures by making circular modelings to obstacles. Based on the actual collision problem of USV in the marine environment, this paper proposed elliptical clustering method of dynamic vessels to improve the obstacle avoidance efficiency. Furthermore, our algorithm also applies the International Regulations for Preventing Collisions at Sea (known as COLREGS, for collision regulations), and USV return the specified routes after the obstacle avoidance. Finally, we shown simulation analysis and experimental testing of avoidance algorithm.