A Bayesian Model to Smooth Telepointer Jitter

Abstract

Cursor prediction is the problem of predicting the future location of a user’s mouse cursor in a distributed environment where network lag is present. In general, cursor prediction is desirable in order to combat network jitter and provide smooth, aesthetically pleasing extrapolation. Gestures can also be difficult to interpret if network jitter becomes too severe.

This paper proposes a Bayesian network model for addressing the problem of cursor prediction. The model is capable of predicting the future path of the cursor while drawing a gesture, in this case an alphabetic character. The technique makes use of Bayesian learning techniques in order to obtain realistic parameters for the proposed solution. The model is then implemented and tested, yielding substantial improvements over previous methods. In particular, the model is at least twice as accurate as a simple linear dead reckoning algorithm run on the same dataset. Furthermore, a by-product of the model is its ability to correctly recognize the alphabetic character being drawn 84% of the time.