| 摘要 |
A moving object detection method with which a region of a moving object is accurately extracted without being affected by a change in shape or size or occlusion of the moving object and in which a distance indicating a similarity between trajectories of an image in each of the blocks included in video is calculated (S203) and a group of similar trajectories is identified as one region based on the distance (S209). Step S209 includes, for each of the thresholds, (1) non-linearization of the distance, (2) calculating a geodetic distance between plural trajectories from the distance after non-linearization (S204), (3) specifying, as region candidates, a group of trajectories which are distant from each other by the geodetic distance of a finite value, from among the trajectories (S205), calculating an interregional geodetic distance that is a scale indicating the similarity between the region candidates (S206), and selecting, from the region candidates, a region candidate having the interregional geodetic distance that satisfies a predetermined condition, as a result of segmentation (S207). |
