Statistical Removal of Shadow for Applications to Gait Recognition

The purpose of this thesis is to mathematically remove the shadow of an individual on video. The removal of the shadow will aid in the rendering of higher quality binary silhouettes than previously allowed. These silhouettes will allow researchers studying gait recognition to work with silhouettes unhindered by unrelated data. The thesis begins with the analysis of videos of solid colored backgrounds. A formulation of the effect of shadow on specified colors will aid in the derivation of a hypothesis test to remove an individual's shadow. Video of an individual walking normally, perpendicular to the camera will be utilized to test the algorithm. First, the algorithm replaces shaded pixels, pixel values determined to be shadows, with corresponding pixels of an average background. A hypothesis test will be employed to determine if a pixel value is a shaded pixel. The rejection region for the hypothesis test will be determined from the pixel values of the frames containing a subject. Once the shaded pixels are replaced, the resulting frames will then be run through a background subtraction algorithm and filtered, resulting in a series of binary silhouettes. Researchers can then utilize the series of binary silhouettes to accomplish a gait recognition algorithm.