Template based classification of multi-touch gestures

We propose a probabilistic classifier for multi-touch gestures specified by users themselves. The template-based gesture classifier allows selecting gesture types more freely without constraints regarding implementation issues and considers multi-finger or bi-manual operations. The statistical approaches to the classification scheme are presented. The basic concepts of separating input into tokens, retrieving local features and applying a new method of sensor fusion under uncertainty are adaptive to broader application ranges. Results from testing against a set of sophisticated samples show that this approach performs well and, while recognition benefits from more complex gestures, it also distinguishes subtly different gestures.

► Statistical classification, formalization/algorithm and empirical proof.
► Concept of dividing gestures into tokens (different sensors).
► Maximum likelihood matching of sensor data.
► Represents special sensor fusion (token features “of unknown sources/sensors”).
► Combining distance and probability measurements for classification.