Real Time Tunnel Based Video Summarization Using Direct Shift Collision Detection

This paper presents a real time tunnel based video summarization using direct shift collision detection. The algorithm first detects objects from each video frame and segments them into slices using HOG object detection. Slices are then tracked as a tunnel which describes movement of objects in time space. We then propose direct shift collision detection algorithm (DSCD) to compute a distance for compacting tunnels. Shifting tunnels using DSCD yields the results of multiple activity tunnels appeared simultaneously while they are originally appeared at the different time. In order to solve such problem, our proposed film map generation technique is used to summarize a video which leaves just-in-time renderer to render only necessary frames. The combination of these three proposed methods reveal an overall performance that gives us real time results without losing the main contents of summarized video.