On-line analysis of music conductor´s two-dimensional motion

The authors propose an online method for understanding a human´s conducting action for a chorus or orchestra observed through a vision sensor. The vision system captured images of conducting action and extracted image coordinates of each endpoint of the baton every 1/15 second. A fuzzy-logic-based inference algorithm was then applied for the information involved in the conducting action. Complementary algorithms are also proposed for identifying first beat, static point, and dynamics of playing. Through a sequence of experiments organized around a system including an IBM PC-386, and a special-purpose vision system and sound module, the algorithm was found to faithfully follow the conductor´s motion in playing a given song