Fast and accurate background subtraction for video surveillance, using an adaptive mode -tracking algorithm
Codrut Ianaşi, Vasile Gui, Florin Alexa, Corneliu Ioan Toma · 2005
Abstract: − Background estimation and subtraction is a critical and time consuming step in moving object segmentation for video surveillance. Nonparametric kernel density estimation has been successfully used in modeling the background statistics, due to its capability to perform well without making any assumption about the form of the underlying distributions. To obtain realtime performance of the nonparametric estimator, we recently proposed an algorithm based on mean shift mode-tracking and a rough histogram test to fast discard foreground pixels from exact evaluation. In the present work, an improvement of the new algorithm is proposed, leading to faster background change tracking capability and more accurate background estimation.