Statistical Approach to Background Subtraction for Production of High-Quality Silhouettes for Human Gait Recognition

This thesis uses a background subtraction to produce high-quality silhouettes for use in human identification by human gait recognition, an identification method which does not require contact with an individual and which can be done from a distance. A statistical method which reduces the noise level is employed resulting in cleaner silhouettes which facilitate identification. The thesis starts with gathering video data of individuals walking normally across a background scene. The video is then converted into a sequence of images that are stored as joint photographic experts group (jpeg) files. The background is subtracted from each image using a developed automatic computer code. In those codes, pixels in all the background frames are compared and averaged to produce an average background picture. The average background picture is then subtracted from pictures with a moving individual. If differenced pixels are determined to lie within a specified region, the pixel is colored black, otherwise it is colored white. The outline of the human figure is produced as a black and white silhouette. This inverse silhouette is then put into motion by recombining the individual frames into a video.