Abstract: Motion feature is one of important clues representing content of a video sequence, and also it is the typical content that distinguished video data from image data. It is an essential aspect and while extracting the comprehensive description of the video content. In this paper we propose a spatial-temporal motion feature based video retrieval scheme. The initial trajectory of balls is obtained, according to the discrete location of the ball. And then the trajectories are modified by the collision point. The trajectory dots sequence is considered as feature of motion, at the same time, we propose a video clip retrieval method by step which is based on the trajectory dots sequence order match: firstly, the trajectory is dived up into several sections, according to the collision point, and then the trajectory segments of the video are re-sampled, finally, the trajectory model matching method is used to achieve retrieval. The experimental results show that the proposed scheme could query exactly and achieve semantic content retrieval to some extent, it provides the retrieval result consistent with the snooker video’s moving trajectory content of people’s subjective understanding.