A Robust Motion Detection Algorithm for Complex Background Using Statistical Models

Abstract

Based on the fact that most of the algorithms assume that the camera is fixed and the changing background is learned in the training period, a robust algorithm is proposed for complex background where a shaking camera, changing background and shadows are presented. It combines a new improved mixture of Gaussians model and a square neighborhood matching algorithm to eliminate shadows and reduce false positive detections caused by camera motion and changing background. Experiments results demonstrate the efficiency and accuracy of this algorithm.