Background modeling adaptive to outdoor illumination variation and foreground detection approach

Considering the appearance of illumination variation in outdoor video surveillance, a real-time background modeling framework, which is also composed of accurate foreground detection, is established. In view of the accuracy of foreground detection, a threshold based on the histogram of pixel's intensity difference between neighboring frames is proposed. On account of the real-time background modeling, a fast estimation approach on parameters of autoregressive model is presented. Considering the adaptability to variable illumination, a texture background model insensitive to outdoor illumination variation is designed. Thus, a uniform model named auto regression and texture (ART) is obtained. According to the established confidence intervals with perturbation of pixel's intensity and its local texture, foreground in scenes with different illumination variations is successfully detected. The experimental results indicate that the framework is adaptive to and can exactly track outdoor illumination variation in real time. Moreover, foreground detection is successfully accomplished.