| With the development of network communication and multimedia technology and the information needs growing, multimedia information becomes a major data source of various types of systems. Of that, the position of video are growing in the network multimedia elements, how to analyze the massive video quickly and efficiently, especially the video of some special scene, become more and more important. Now, the difficulty of network video is that the video data is very large, the video scene is rich, the objects in video are so many, and how to analyze the event happens in the video is more difficult.TRECVID, as the most authoritative international evaluation on the video retrieval, has begun to focus on these high-potential research directions. The research topic of this paper comes from the event detection of this evaluation, mainly focuses on opposing flow event detection from the airport video, proposes a no training and efficient real-time detection method, based on moving analyzing.Based on the moving detection with frame difference, this paper uses the image local feature, corner point, to be the feature of the moving object, and use optical flow algorithm to acquire the prediction of moving direction and speed. Generally considering the result of moving analyzing, we can detect and calculate the time opposing flow event happens by judge queue.Experiment results show that, this system has a good processing speed to ensure the real-time in opposing flow event detection, and the detect results also reach the average level of TRECVID evaluation. In addition, this paper also adds the laboratory scene video for comparison test, and the test results have big substantial promotion, and the detective speed of the system also improve. Finally, the method this paper proposed, avoid the large-scale video data training, and can guarantee a good detection results at the same time, it has a high research value for real-time detection and analysis of the similar event. |
PDF Full Text Request