Radar-based gesture interaction: from sensing to recognition

In the recent years, radar-based gesture recognition has seen an increase in popularity. Indeed, radar sensors have various advantages over the widely used wearable and image-based sensors. They are less intrusive and less sensitive to ambient and lighting conditions. Moreover, as the radar waves are able to propagate through materials depending on its permittivity, radar sensors are the go to candidates to perform gesture recognition through materials. In this work, an off-the-shelf radar, the Walabot, is used to acquire a dataset of hand gestures performed in front of three different material plates (wood, PVC and glass) placed in-between the radar and the participant. In order to evaluate the suitability of using the Walabot in such scenarios, multiple recognizers based on template matching are trained using the acquired hand gesture dataset and their accuracy at performing hand gesture recognition is evaluated. This thesis presents the complete methodology followed to perform hand gesture recognition through 3 different material plates (wood, PVC and glass), from the acquisition of the dataset using the Walabot and the processing of the radar data with the RadarSense pipeline, to the training and testing of the recognizers with the QuantumLeap framework.