Background Subtraction from Auto-Exposure Surveillance Camera in Restaurant

This paper introduces an approach to background subtraction from auto-exposure surveillance camera. The proposed method adapts the concept of codebook while the original codebook has a limitation in auto-exposure camera. These videos have light adjustment over time and it leads to brightness variation, and the variation could occur globally or locally. The proposed method uses the advantage of YCbCr color space and adjusts the criteria for the segmentation in the original codebook. Moreover, the proposed method can process the input video in real time and can reduce noises due to illumination change. The experimental results show that the proposed method can reduce a lot of noises in the video, and also have a better accuracy that the original codebook method.