Abstract
In this paper, we describe a simple but effective and fast video activity detection algorithm to provide an efficient pre-processing tool for video camera feeds. It also presents a range of applications, including post-production of TV camera video streams, analysis of surveillance videos, visual scene classification and camera control for automated recordings. Given a video sequence, our algorithm detects moving pixels by modelling their differences in Gaussian distribution through online estimation of their statistics to determine an adaptive threshold for final classification of active or static scenes. The originality of our contribution can be highlighted as: a powerful activity detection tool inside videos to facilitate efficient visual content analysis; online estimation of statistics and adaptive determination of a threshold for automated classification of active or static scenes. Extensive testing supports that the proposed algorithm achieves excellent performances.