Background Subtraction and Frame Difference Based Moving Object Detection for Real- Time Surveillance

摘要

A new real-time algorithm is proposed in this paper for detecting moving object in color image sequences taken from stationary cameras. This algorithm combines a temporal difference with an adaptive background subtraction where the combination is novel. When changes occur, the background is automatically adapted to suit the new conditions. For the background model, a new model is proposed with each frame decomposed into regions and the model is based not only upon single pixel but also on the characteristic of a region. The hybrid presentation includes a model for single pixel information and a model for the pixel's neighboring area information. This new model of background can both improve the accuracy of segmentation due to that spatial information is taken into account and saliently speed up the processing procedure because portion of neighboring pixel can be selected into modeling. The algorithm was successfully used in a video surveillance system and the experiment result shows it can obtain a clearer foreground than the single frame difference or background subtraction method.