Air+Touch gesture recognition : algorithms, software, and experiment

Human Computer Interaction became omnipresent in recent years, especially gesture recognition. Thanks to major advances in machine learning, gesture recognition has become an everyday tool. Currently, almost everybody has mastered 2D gestures with the use of smartphones. However, limits of these 2D gestures have emerged which has led researchers to focus on 3D gesture recognition. These two types of gesture enable to create Air+Touch recognition. It simply consists in an environment where gestures can be 2D or 3D. We have decided to deal with Air+Touch gesture recognition for this thesis. There is a clear lack of research on this subject. A problem related to this lack is that devices that can sense and retrieve data for the two types of gestures, are quite rare. The 3DTouchpad device can sense Air+Touch gestures. Therefore, we can build our own gestures set based on this device and on all the Air+Touch gestures gathered from the literature. Once it was defined, we integrated the 3DTouchpad inside the QuantumLeap framework where we tested all their different recognizers to see which one fits the best with Air+Touch gestures. Each test was performed according to two scenarios: User-Independent, User-Dependent. We also analyzed two gestures elicitation studies based on the 3DTouchpad to confirm that our Air+Touch assumptions were corresponding to 60 participants feelings. To conclude, we are going to discuss about the benefits and contributions of our work before suggesting some future promising works.