Hand gesture video browsing for broadband-enabled HDTVs

This paper presents three methods for hand gesture detection and recognition that can be applied to online video browsing. These methods aim at recognizing hand signs and positions using a single webcam, which can in turn, be used to control a broadband-enabled HDTV. The hand gesture can be trained to suit the user preference. We first provide an analysis of pattern matching, histogram back projection, and the use of Fourier's descriptors. These methods achieve good reliability and acceptable resource consumption. We compare these methods with a new method based on H.264 motion vectors that directly analyzes video in the compressed domain. It will be shown that this technique provides a faster and accurate way to recognize motion trajectories that may correspond to letters or alphabets. The extracted gesture or trajectory information can then be used for various multimedia applications, including improving human-TV interaction.