Key frame is a simple yet effective form of summarizing a long video sequence. We present a novel key frame extraction algorithm. The number of and where to place the key frames are determined by the Perceived Motion Energy (PME) feature of a shot. The motion pattern of a typical scene is composed of a motion acceleration process and a following deceleration process and this pattern is repeated in video sequence. A triangle model is developed to describe the motion pattern and segment video sequences. The frames at the start point and end point of motion acceleration are selected as key frame candidates. So one can infer the movement in between two key frames and the salient action content in video can be abstracted by these key frames.