Night Video Surveillance Based on the Second-Order Statistics Features

Night video surveillance is crucial to construct an all-weather video surveillance system. However, night video surveillance faces several problems: no color information, low brightness, low contrast, and low signal to noise ratio (SNR). These problems can introduce serious false and missing object detections. In this paper, we propose a novel night video surveillance method based on the image second-order statistics features to overcome the aforementioned problems. First, the block-based foreground detection and dual foregrounds fusion methods are used to extract the candidate object regions. Second, by extracting the second-order statistics features in the candidate moving object regions, we may identify the pedestrians via the support vector machines (SVM). Experimental results show that the pedestrian detection and recognition accuracy with the near infra-red (NIR) images can approach 93% and the system efficiency is 25 fps.