3D Gesture recognition: an evaluation of user and system performance

We report a series of empirical studies investigating gesture as an interaction technique in pervasive computing. In our first study, participants generated gestures for given tasks and from these we identified archetypal common gestures. Furthermore, we discovered that many of these user-generated gestures were performed in 3D. We implemented a computer vision based 3D gesture recognition system and applied it in a further study in which participants used the common gestures generated in the first study. We investigated the trade off between system performance and human performance and preferences, deriving design recommendations. We achieved 84% recognition accuracy by our prototype 3D gesture recognition system after tuning it through the use of simple heuristics. The most popular gestures from Study 1 were regarded by participants in Study 2 as best matching the task they represented, and they produced the fewest recall errors.