Foreground object detection under camouflage using multiple camera-based codebooks

Video surveillance is used to monitor activities of people and to take necessary actions when an abnormal activity is noticed. The initial step of video surveillance is object detection, which is done by background subtraction. In this paper, we present a multiple camera-based object detection method under camouflage i.e., when the background color is almost similar to the foreground color. In our proposed method, the pixels corresponding to the background sample values are quantized into codebooks which are sufficient to represent a background. Codebooks are extracted for each of the cameras. Subsequently, they are combined to extract the foreground from the background. Finally, the disparity map extracted from the multiple views is used for proper identification of the foreground when there is confusion between the foreground and the background due to color similarity. Experimental results show that the proposed method is quite robust to camouflage.