markers-color probabilities the classi¯er is able to ef-
fectively cope with considerable illumination changes,
and therefore it is able to robustly track colored mark-
ers even when there are cluttered and dynamic back-
ground conditions. We utilize ARTag (Augmented
Reality Tag) [3] to compute the extrinsic parameters
for estimating the guitar position. Then we calibrate
the cameras for calculating the projection matrix at
online process. To assist the recognition of correct
chords while the guitar is moving, we de¯ne the world
coordinate on the guitar neck as the guitar coordinate
system. We utilize a triangulation [4] technique to es-
timate 3D position of the player's ¯ngers using stereo
cameras, and then recognize if a guitar string is be-
ing pressed or not. Next, we apply PCA (Principal
Component Analysis) [5] to reduce the input dimen-
sion which allows each guitar chord to be classi¯ed
more accurately. As a result, players can recognize
the guitar chords they are playing during the song in
real time.
Obviously it would be of great assistance to learn-
ers if they are able to recognize the chord being played
by the left hand. The system improves accuracy be-
cause it can identify whether the ¯nger positions are
correct according to the ¯nger positions required for
the piece of music they are playing (i.e., the gui-
tar teacher inputs the correct ¯ngering for each gui-
tar chord and these are then incorporated into the
system that we have provided). Applying the pro-
posed method, beginners can automatically identify
whether their ¯ngers are in the correct position which
makes this a guitar application. The proposed teach-
ing system would be invaluable to teachers and have
a wide application to supporting people learning to
play the guitar.