Abstract
It is important to divide video data semantically for retrieving it efficiently. It is desirable to divide it in consideration of camera motions. Because of this, video creators often put their minds into their camera motions. Therefore, in this paper, we propose a method of detecting camera motions and dividing video data. In camera motion analyses, there are two problems: camera vibration and foreground object motion. In our method, to avoid the effect of camera vibration, we used spatio-temporal images for detecting camera motions. Also, to avoid the effect of foreground object motions, spatio-temporal images in consideration of composition of video data.