If we get a region that may contain a candidate category, we apply object category model to detect an object category in this region. There will be many detection results with different location and different scale in this region. We only consider the detection results as valid when the overlay of the returned bounding box R and the motion region Q is greater than 50%. Among these valid detection results, we take the one with maximal score as the detection result for this region. It contains the returned bounding box and the corresponding detection score. For every considered object category model, we will obtain a detection result like that.