A Personalized Video Synopsis Framework for Spherical Surveillance Video

Abstract

Video synopsis is an effective way to easily summarize long-recorded surveillance videos. The omnidirectional view allows the observer to select the desired fields of view (FoV) from the different FoV available for spherical surveillance video. By choosing to watch one portion, the observer misses out on the events occurring somewhere else in the spherical scene. This causes the observer to experience fear of missing out (FOMO). Hence, a novel personalized video synopsis approach for the generation of non-spherical videos has been introduced to address this issue. It also includes an action recognition module that makes it easy to display necessary actions by prioritizing them. This work minimizes and maximizes multiple goals such as loss of activity, collision, temporal consistency, length, show, and important action cost respectively. The performance of the proposed framework is evaluated through extensive simulation and compared with the state-of-art video synopsis optimization algorithms. Experimental results suggest that some constraints are better optimized by using the latest metaheuristic optimization algorithms to generate compact personalized synopsis videos from spherical surveillance videos.