Abstract:

Focusing on the precise detection of moving objects, we propose a new local texture descriptor—LBP_Center, which is robust to noise and invariant to gray scale. And together with the color information of pixels we apply it to the background model. Then we update the model by random sampling and introduce background complexity to remove the noise of the multimode dynamic background. Experimental results on standard testing datasets indicate that the new method has good robustness to soft project shadows and slow illumination change, and better comprehensive performance in comparison with other algorithms.

Key words: detection of moving objects, background subtraction, local texture feature, soft projected shadows