A Foreground-Background Segmentation Algorithm for Video Sequences

Abstract

Moving objects extraction is a crucial part of video surveillance system. This paper presents a foreground-background algorithm for motion detection. It is based on traditional adaptive mixture Gaussian model. By dynamically adjusting the parameters and the number of Gaussian components, the computation cost reduced greatly. In order to solve detected moving target based on Gaussian mixture model easily broken, two-way matching method on the basis of frame difference thoughts with a series of image filtering methods are combined. In a stable outdoor detector, the algorithm deals with lighting changes, swaying of leaves, and various noises reliably. The proposed algorithm can identify moving objects more exactly than traditional method. An intrusion detection alarming system can be built to discover the abnormity by using the algorithm to process and analyze the video sequences.