Net Detection Of Soccer Video

Video Event Retrieval is to infer the video event and use it as the content to perform the video retrieval. Here the event is defined as the medium-level spatiotemporal entity interesting to users having, certain context cues corresponding to the specific domain knowledge model. This paper discusses mainly the net detection as one of context cues to infer the soccer video event. The net detection algorithm includes the following four steps mainly: 1. select the appropriate color space and perform the binary segmentation based on the white pixel; 2. determine the representative connective image; 3. perform the connectivity enhancement based on dilation; 4. verify the net detection result based on grids' statistic. To determine the thresholds involved in the algorithm the training method is adopted. After the threshold training the net detection experiments are performed to evaluate the algorithm performance. Experiments show that the proposed net detection algorithm is effective.