Department of Electrical and Computer Engineering, Qazvin Branch, Azad University, Qazvin, Iran
Generally human vision system searches for salient regions and movements in video scenes to lessen the search space and effort. Using visual saliency map for modelling gives important information for understanding in many applications. In this paper we present a simple method with low computation load using visual saliency map for background subtraction in video stream. The proposed technique is based on finding image segments whose intensity values can be distinguished accurate. The practical implementation uses a sliding window approach, where the distributions of the objects and surroundings are estimated using semi-local intensity histograms. This introduced method requires no training so it can be used in embedded systems like cameras due to low load in calculation. So with our background subtraction algorithm we can detect pre-defined targets. Also the automatically video regions detected by proposed model are consistent with the ground truth saliency maps of eye movement data. Comparisons with state-of-the-art background subtraction techniques indicate that the introduced approach results in high performance and accuracy.