Real time trajectory search in video summarization using direct distance transform

Abstract

Searching for human trajectories in a summarized video is a way of analyzing such video within a smaller group of human moving along the same path. We propose a real time novel method for similarity search for human trajectories using a distance transform of each extracted human tunnel. We integrate this method with our previous work, Real Time Tunnel Based Video Summarization using Direct Shift Collision Detection (DSCD), which provides the human tunnels with trajectory information for being analyzed. The algorithm first creates a distance transform from provided trajectory information. A model of trajectory for screening irrelevant tunnels out of the summarized video is then used as a query. An efficient technique as in DSCD is used for calculating a direct distance transform (DDT). Then a similarity between trajectory model and human trajectories are ranked. The advantage of linear time complexity of both DSCD and distance transform gives us a real time search results while more relevant summarization output can be obtained.

From video summarization to real time video summarization in smart cities and beyond: A survey

“…In an unsupervised manner, a deep summarizer network is used to reduce the distance between training films and the distribution of their summarizations. A summarizer like this can then be used to estimate the best synopsis for a new video (Cooharojananone et al, 2010 ).…”