Compressed-domain Video Synopsis via 3D Graph Cut and Blank Frame Deletion
Wenjuan Liao, Zhigang Tu, Shizheng Wang, Yongzhou Li, Rui Zhong, Hui Zhong · 2017
With the wide adoption in surveillance video analysis, some pixel-domain analysis methods are not only used in academia but are also garnering the attention of industry. However, because of the increasing data volume of video, how to analyze and browse surveillance videos in a fast and effective way has continued to present an intractable problem in practical applications. Therefore, a synopsis analysis method based on blank frame deletion in the compressed domain is proposed in this paper to achieve fast synopsis analysis and video browsing. Our main contributions are: 1) An algorithm of compressed-domain 3D graph cut is introduced to support the fast object tubes segmentation and synopsis analysis in the compressed domain. 2) A concept, referred to as virtual blank frame, is introduced and applied into synopsis analysis. This proposed video synopsis method can further synopsize the video in the spatio-temporal domain through deleting the virtual blank frames after real blank frame deletion. 3) A comprehensive solution, synopsis analyzing based on blank frame deletion in the compressed domain and fast browsing synopsis video in the pixel domain, is provided for video synopsis, which is effective and efficient for video surveillance. Experimental results show that our approach maintains a stable performance in fast synopsis analysis in the H.264 compressed domain.