Gesture Recognition to Control a Moving Robot With FMCW Radar

Gesture recognition with radar sensors experienced great scientific interest in recent years and may become an alternative or supplement technology to camera based computer vision. However, most publications focus on static radars, which is not applicable in case of moving robots, where the radar is attached to the robot itself. This paper presents a system for automated radar data acquisition and labeling in context of radar based gesture recognition, mounted on a moving robot. It is used to collect labeled radar data of 4 different gestures to control the robot. In addition, results of a comparably small CNN, which processes range-Doppler- and range-angle-images, achieving 86 % of evaluation accuracy is presented.