CONCLUSIONS
In this paper, a vision-based method to recognize
the guitar chord played by the left hand has been
presented. We detect the position of a player's ¯nger
markers by using on-line adaptation of color proba-
bilities and a Bayesian classi¯er which can deal ef-
fectively with considerable illumination changes and
a cluttered background. We utilize ARTag's marker
information to track the guitar neck, therefore the
system can produce the correct chord output while
players are moving the guitar, and at the same time
it can cope with ARTag occlusion robustly. The tri-
angulation method is utilized to compute the 3D posi-
tions of ¯ngers to recognizing whether a guitar string
is pressed. We use the results of the guitar and ¯n-
ger detection to compute the guitar chord by utilizing
the PCA method for more accurate results. By ap-
plying this methodology to di®erent conditions, it has
been experimentally demonstrated that the proposed
methodology gives robust and accurate results even
when used in real time.
The proposed method would be of assistance to the
guitarists by providing real time feedback by check-
ing it against the preloaded accurate ¯nger positions
required by each musical piece. As a result, it would
allow the player to gain a higher level of enjoyment
during the lesson. As future work, we are planning
to further re¯ne the problem of the ¯nger markers
by removing these markers which may result in even
greater user friendliness.