Real-time video surveillance summarization and trajectory search using direct shift collision setection

Abstract

Reviewing on video surveillance contents for security monitoring is a time-consuming task. A novel real-time video surveillance summarization framework intended for minimizing time usage for time critical tasks is presented, based on shifting of object tunnels in time-space. Tunnel is proposed as an individual structure of time-dimensioned object. In order to summarize an endless video into a shorter duration with customization and an extension for understanding of individual object, this research proposes three real-time algorithms. Direct Shift Collision Detection (DSCD) has been implemented for extremely fast shifting tunnels together in time-space. A summarized video can be customized by many approaches. An early trajectory searching is applied with the same technique of our DSCD. Direct Distance Transform (DDT) instantly gives trajectory simi-larity between tunnels and users query. The most important step for identifying each individual object is background subtraction. Dynamic Region Adaptation (DRA) is proposed as another background subtraction algorithm aimed to select the best objects foreground before making a tunnel. DRA also helps DSCD to summarize a video more accurately. The proposed framework is able to respond the results by real-time performance approaches without losing the major events of original video stream.