CENE ANALYSIS AND VIDEO TAGGING

Abstract

Computer scientists have made never-ending struggles to reproduce perceptive video understanding abilities of human minds onto automated vision systems. There exists a surge in research and studies in the domain of autonomous activity recognition and the detection of uncommon and unusual events, with the advancement of technology and specifically video surveillance cameras. However, analysis of video content in public or crowded scenes continued to remain a challenging task due to core difficulties and issues such as severe and complex occlusion among objects in a dense scene and low quality of footage recorded during surveillance. Besides, it is uncommon to get vigorous detection of unusual and absurd events, which are vague, rare and can easily be confused with noise. This project provides solution for resolving confusing visual observations and overcoming the issue of uncertainty and unreliability of conventional methods of activity analysis by using a visual setting with for surveillance camera and the development of an automated system for scene analysis using video tagging. It proposes a structure to perform video analysis in detecting suspicious and doubtful activity within the large and tremendous amounts of real time video data that occurs in today’s world of ubiquitous surveillance video.