The following videos demonstrate a particle filter-based baton motion prediction system, developed by Donald Dansereau, and described in our recent paper, Predicting an Orchestral Conductor's Baton Movements Using Machine Learning. The results show effective prediction of up to 250 ms in real time.
The experimental evaluation task, carried out by Nathan Brock, the conductor who took part in this research, is available here.
| The templating process. The white dot shows the real input trajectory, the gray lines and dots show a generic template tracking this input, and the red line/purple dot show the adapted template being formed. | ||
| Performance, showing a prediction range of 250 ms (playback is slowed). The white dot again represents the input trajectory, the large purple dot represents the tracked position on the template, and the smaller white/green dot shows the predicted baton position. | ||
| This is a performance showing only the input data (large white dot) and prediction (smaller green dot). | ||
| This is the same performance as above, shown in realtime, and with accompanying audio. |