A Background Model Estimation Algorithm Based on Analysis of Local Motion for Video Surveillance

Knowing the background model of a video scenario simplifies the problem of object segmentation and object tracking in the automated video surveillance applications. In this paper, a new algorithm for background model estimation was presented, which is useful in situations where an unobstructed view of the background is not always available. Discovering the true background interval in pixel's intensity history through local analysis of motion and spatial information, it avoids the problems of blending pixel values present in many current methods, such as mean filter and Kalman filter. Experimental results of applying our approach on a sequence of an indoor scene are provided to demonstrate the effectiveness of the proposed method