Generalised Spatio Temporal Feature Based Important Activity Synopsis Generation

Traditional video analysis methods can generate summary of day's long videos. While generating synopsis video maintaining the motion structure of important activities present in a video sequence is of great concern in research communities and industry. In this paper, we present an automatic and scalable approach for automatic detection of important activities in a video based on different spatiotemporal criteria used as a signature. To maintain the important context cues we propose an online motion structure preserved synopsis approach, which can retain the behavior interactions between different objects in an original video while condensing as much content as possible. A hierarchical fashion is employed to efficiently search important activities present in a video sequence, and generating synopsis video of those important activities in which both the spatial collision and the temporal consistency are considered. Experimental results on numerous (six) video sequences demonstrate the promise of the proposed approach.