ViBeExt: The extension of the universal background subtraction algorithm for distributed smart camera

This paper presents a technique for improving the accuracy and robustness of the high speed background subtraction method called ViBe at very limited additional cost. Despite the through-output efficiency of ViBe, which is suitable for embedded in smart camera, it also incorporates several innovative mechanisms differing from those classical algorithms, like GMM, such as propagated current pixel value into the background model of a neighbouring pixel. However, such propagation neglect the difference of neighbouring pixels, which will introduce some defects. In this paper, not only we introduce the edge characteristic info for more accurately updating the background model, but also through neighbourhood mean filter to strengthen the pixel's spatial correlation, which is the primary principle behind ViBe. Thoroughly experiments have been carried out, and efficiency figures show that our method improved the accuracy and robustness of ViBe with a little additional cost, which is more suitable for distributed smart camera.