Design of moving object detection schemes for challenging surveillance environments

Abstract

A multitude of Moving Object Detection (MOD) algorithms have been put forward in newline newlinethe last few years for addressing several challenges affecting object detection perfor- newlinemance in surveillance videos. This is in view of the rising demand and significance newline newlineof real time applications proposed by the computer vision and video surveillance newlinecommunities. The fundamental procedure for the computers for scene understanding newlineand further decision making is the accurate segregation of the moving objects called newline foreground (FG) from the ideally static information called the background (BG) . newlineBasically such an MOD method functions well when three conditions are met, i.e. newlinestatic camera, constant background and constant illumination. Many traditional newlineas well as contemporary algorithms are developed to implement MOD with these newlineassumptions. The problem with these methods becomes severe when it comes to newlinethe case of real object detection systems where the above three conditions are rarely newlinesatisfied. To be more specific, the degraded performance of these existing MOD newline newlinemethods occur mainly due to some unaddressed challenges such as dynamic back- newlineground, noise, incomplete/missing pixels, subsampled data and low resolution video newline newlinedata. Accordingly these videos must be preprocessed before using it for higher level newlineapplications like object tracking, action recognition, etc. newlineThe increase in dimensionality of the modern video data is a clear hindrance newlinefor the video processing tasks. Even though such defects are well addressed for newlineone dimensional signals (e.g.: speech) and two dimensional signals (e.g.: images), newlinethe restoration of defects appearing in three dimensional signals (e.g.: video) are newlinenot resolved perfectly. The inherent redundancy present in natural videos is a newlinesignificant prior, in fact videos posses considerable temporal redundancy. newline