Two Hand Tracking Using Colour Statistical Model with the K-means Embedded Particle Filter for Hand Gesture Recognition

Particle filtering is an efficient and successful technique for tracking 2D and 3D motion through an image. We present the enhanced tracking of two hands based on a statistical model using only a skin colour feature with particle filtering for gesture recognition. Our framework employs one particle filter per hand individually with the pixel-wise classification of the likelihood of the skin in the window search. The skin classifier decision was trained from a set of skin samples in YCrCb space using an elliptical model. The tracking scheme employs a reliability measurement derived from the particle distribution which is used to adaptively weight the colour classification. The K-means algorithm is used to discriminate the split and merge between left and right hand. Experiments with a set of videos including the movement of two hands in cluttered backgrounds show that adaptive use of our scheme provides improvement compared to use with other techniques such as mean-shift tracking.