Sequential Monte-Carlo Estimation of Background Image for Background Subtraction under Changing Illumination

M. Tsuchida, T. Kawanisi, H. Murase, and S. Takagi (Japan)


tracking, background subtraction, background estimation,illumination change, sequential Monte-Carlo estimation.


This paper proposes a background image estimation method for background subtraction under changing illumination. In our method, a background image is generated as a weighted linear combination of the prepared background images. The weights change slowly as time advances. We therefore adopt the sequential Monte-Carlo method to track the weights, and background images are generated. A generated image in which area having pixel value equal to that of the input image is the largest is selected as the background image under the illumination. In the experiments, background estimation and subtraction were carried out under changing illumination conditions; i.e., geometrical setup, brightness, and color. The results show that background images are correctly estimated even if 80% of the area of the input image is occupied by a foreground object.

Important Links:

Go Back