Efficient Tracking of Pianists’ Finger Movements

Abstract

Pianist finger movement contains information about note preparation, accent information as well as personal identifiers. However, no system for detailed finger tracking presently exists that can also be used ecologically: at an acoustic piano whilst also incurring acceptably low disruption to the performance. An optical motion capture system is described using a monocular camera setup to track passive UV paint markers placed directly onto pianists’ fingers. Tracking accuracy and 3D estimation algorithms are evaluated. This low-cost, efficient system maximizes its uses in performance analysis experiments which are most importantly not confined to laboratory environments.

Acknowledgements

This work was partly supported by an EPSRC Doctoral Training Grant for the first author. The authors would also like to thank Tom O’Hara at the University of Glasgow for his help in designing the camera apparatus.

Notes

¹This particular measurement is stated by Rumelhart and Norman (1982) when investigating speed of typists’ finger movements, however, other statements on the rate of keystrokes a pianist can make are found in Smith (1953) who counted the number of keystrokes pianist Simone Barrere used during a performance of Schumann’s C major Tocatta Op. 7. By dividing the total time of the performance by the number of keystrokes, Smith calculated that the pianist was playing at a rate of 24.1 notes per second. However, the rate at which a pianist can make consecutive keystrokes depends also on the complexity of the music. A fast trill between two notes (particularly when using alternating hands for each note) may be played faster than this 24 notes per second. Overall, through many other examples provided in Smith (1953) the consensus is that the upper speed of fingering ranges from 20--30 notes per second.

²By resting position, we aim to have the hands resting flat on the keyboard with the fingers stretched out. This is so the software can estimate the initial depth of each point as approximately the distance between the piano keys and the camera lens. Obviously this will not be entirely accurate, but the aim is to retrieve a starting depth for each point from which consecutive depths can be measured.