Human-Computer Interactions Using Low-cost Millimeter Wave Radar Sensors

With recent developments in intelligent sensing hardware and algorithms, the performance of smart home applications for human-computer interactions through capturing and processing human information has improved significantly. However, the data security and privacy of camera based sensing techniques are often overlooked. On the other hand, newly-developed commercially available millimeter-wave radar sensors are small in size yet powerful enough to achieve imaging capabilities in the form of high-resolution point clouds. In this paper, we explore the possibility of intelligent sensing for human-computer interactions using low-cost off-the-shelf millimeter-wave radar sensors. Particularly, we focus on typical key tasks such as face recognition and activity recognition, using a deep learning based framework for millimeter-wave radar point cloud data. We further design a prototype real-time human-computer interaction system to demonstrate typical scenarios such as outdoor intelligent access control and indoor intelligent monitoring. The experimental results indicate that we achieve an overall average accuracy of 98% in face recognition, while we achieve an average of 98.4% accuracy for classification and a 97.6% accuracy for identification in activity recognition.