Improved Gaussian Mixture Model for Moving Object Detection

Detection of moving objects in image sequence is a fundamental step of information extraction in many vision applications such as visual surveillance, people tracking, traffic monitoring. Many background models have been introduced to deal with different problems. Gaussian mixture model is considered to be one of the most successful solutions. It is a robust and stable method for background subtraction. It can efficiently deal with multimodal distributions caused by shadows, swaying trees and other knotty problems of the real world. However, the method suffers from foreground objects bending into the background too fast. In addition, it can not deal with the problem of slow-moving objects. In this paper, an efficient method is presented to deal with the problem through improvement on the background updating period using different learning rates for the estimated background and foreground pixels. The experiment result shows the method works better than the typical Gaussian mixture model.