Retrieve video clips using the global motion information

Retrieve video clips using the global motion information Tianli Yu and Yujin Zhang Department of Electronic Engineering, Tsinghua University, Beijing 100084, China In this paper, a new scheme is proposed to extract the global motion model from general video sequence. Since global camera motion often conveys the semantic meaning of the video’s content, these global model parameters are used as features for video retrieval. Experimental results show our scheme is quite effective in locating video segments with similar motion content in football game sequences. Introduction: In general video sequences, the movement of the camera always follows the place where the important activity goes on. This means the camera motion contains cues of the semantic meaning of the video, especially in some constrained domain. One good example of this is the football game. In football game video, the camera’s movement is often focusing on the trajectory of the football and the position of the players. This information could be used as the feature for the retrieval of the football sequence. In football games, color and texture feature based retrieval methods will generally failed since the scene won’t change too much from the point of view of color and/or texture. Cameral motion based feature could be a good alternative. The advantage of using global camera motion feature is that it could be extracted easily and robustly. In this paper we propose a simple method to estimate the bilinear motion model parameter of the global camera motion and use the extracted global motion model as the feature for the retrieval of video clips. Global Motion Estimation: The motion of pixels in two consecutive video frames can be divided into global motion and local motion. Global motion is caused by the movement of camera and could be calculated by 7estimating several model parameters using the linear estimation methods [1]. We choose the bilinear motion model and least square estimation to estimate the global camera motion. According to the bilinear model, the global motion vector ) , ( v u of a point ) , ( y x could be expressed as: