Vision-based guitarist fingering tracking using a Bayesian classifier and particle filters

Abstract

This paper presents a vision-based method for tracking guitar fingerings played by guitar players from stereo cameras. We propose a novel framework for colored finger markers tracking by integrating a Bayesian classifier into particle filters, with the advantages of performing automatic track initialization and recovering from tracking failures in a dynamic background. ARTag (Augmented Reality Tag) is utilized to calculate the projection matrix as an online process which allow guitar to be moved while playing. By using online adaptation of color probabilities, it is also able to cope with illumination changes. © Springer-Verlag Berlin Heidelberg 2007.