Adaptive random-based self-organizing background subtraction for moving detection

The adaptability plays a significant role in moving detection. The diverse scenarios in real world still challenge this problem. Therefore, in this paper, we proposed an adaptive moving detection method, namely Adaptive Random-based Self-Organizing back- ground subtraction (ABSOBS) method. This method can adaptively extract the moving objects in various conditions and eliminate the “ghost” pixels simultaneously. Therefore, a robust initialization strategy is proposed to remove the noise pixels caused by the initialized frames. The proposed method uses a random- based scheme which allows the foreground pixels to up- date the neural network with a small probability. This strategy allows our algorithm to efficiently handle scene changes. Moreover, a foreground filter based on random rule is designed to eliminate the “ghost” pixel. More importantly, ABSOBS adopts a regulator to control the updating rate in different conditions. It makes our method easy-to-used and need not to set the parameters manually. The experiment results on various scenarios show that our method improves the detection accuracy for the SOBS and outperforms other state-of- the-art methods.