Adaptive background update based on mixture models of Gaussian

Abstract

In computer vision system, detection of moving targets has interference in subsequent processes including classification, tracking and recognition. Background subtraction method is commonly used in image segmentation for moving region of video. This paper puts emphasis on background model update based on mixture models of Gaussian in complicated situation, and implements an adaptive learning method to update background models. Each pixel is classified into 4 different types: still background, dynamic background, moving object, temporary still object. And the proposed method reduces the computational complexity.