A Method Of Key Frame Extraction For The Surveillance Video

Surveillance system has been widely used in bank, traffic and other fields now. The massive video data places huge burden on the storage space, and people are less satisfied with the traditional system that it only carries on the simple function. How to manage and store the video data effectively is an important research direction in the current intelligent information management. The key frame technology is a good method for it.This paper proposes a novel method of key frame extraction for the surveillance video, combined with the characteristics of the data. According to the large amounts of video frame sequences containing pure background, surveillance video is divided into the key video segments and the redundant video segments. Then we extract the key frames from the key video segment which contains moving objects.Comparing with the current commonly used methods, the background subtraction method is used to detect the moving target in this paper. To prevent the moving object dirtying the initial background, the improved multiple frame average method is applied to initialize the initialization background model. And in order to simplify the Gauss Mixture Model from both robustness and speed, the number of the background model is fixed at two. Judging whether the pixels change, we also check the neighborhood pixels, besides the current pixels themselves. The pixels in per frame do differential treatment respectively with the corresponding pixels and the adjacent pixels in the background. When the results are all bigger than the threshold value, we determine that the pixels are the motion pixels. After that the further process is carried on to reduce image noise.After the division of the video, we extract the key frames from the key video segment which contains moving objects.In this paper, comparing with the current key frame extraction technology, the key frame extraction is based on mutual information, where the mutual information is proposed as the feature parameters between the video frames. The results show that the method can extract the key frames from the surveillance video with less redundancy.