Extracting a Human Area by Background Detection

This paper proposes a technique for extracting a human from a video image stream by detecting the background in the image. Unlike the existent background subtraction method, the proposed method detects the background first without interrupting those people who cross in front of the camera. This is achieved by applying a median filter pixelwise to successive image frames. As we assume the use of a mobile camera, image mosaicing is performed in advance to the median filtering to extract common image areas among successive image frames. Once the background is obtained, it is subtracted from the present images, yielding the images containing only humans or moving objects. Further in the technique, mean shift is employed for detecting a human who has stood still. An experimental result shows satisfactory performance of the proposed method