A Novel Approach for Radar-Based Human Activity Detection and Classification

In this paper, we propose a novel approach for real-time radar-based human activities detection and classification. In this approach, first, the radar transceiver is mounted on the room’s ceiling leading to considerable variations of the relative received power as the subject perform the different activities. Second, to exploit the different activities’ dynamics (i.e. evolution in the time domain), radar data is registered over a long period (around 8 seconds) leading to distinctive signatures corresponding to the different activities. Then Machine Learning (ML) techniques are used for these activities’ classification. The obtained results demonstrate that this approach performs well both in millimeter Wave (mmWave) and in the sub-6GHz bands. We even obtained better results in the sub-6GHz band with an average classification accuracy of 95.7% compared to 89.8% obtained in the mmWave band.