Provided is a technique for extracting information with which it is possible to track an object to be tracked, even if it happens that the object to be tracked is hidden or the like. This image processing device is provided with: a moving region identification unit which identifies, in a video, the image region associated with a moving object shown in the video; a stationary region identification unit which identifies, in the video, the image region associated with a stationary object shown in the video; and an extraction unit which extracts a feature of a partial image that is included in the image region associated with the stationary object, and that does not overlap the image region associated with the moving object.