Object detection and segmentation in sport video

This paper approaches the object detection and segmentation as the first step to semantic object based video retrieval. It proposes an efficient method to segment objects such as athletes and field in sport video by analyzing the information of visual and motion features. Canny operator and morphology method are exploited to detect the field of sports and the result is verified by dominant color calculation histogram. These indicating lines in field are removed by Hough transform and a threshold mixed method. Motion areas in sport filed are detected and segmented by the algorithm. Experiments show that the algorithm performs very good results.