3D-Skeleton Estimation based on Commodity Millimeter Wave Radar

The extraction of human pose from the 3D point-cloud data is an important research challenge in the field of human behavior recognition, which will play a vital role in smart home, elderly-care, gaming, etc. Among the variety of technologies proposed, mmWave sensor based technology has been attracting more attention in the behavior recognition due to its advantages in privacy protection. However, previous studies have focused on using sophisticated mmWave radars (which are typically costly), whereas commodity mmWave sensors have not received much attention due to their low spatial resolutions. In this paper, we propose a skeleton estimation method based on commodity millimeter-wave radar. Through data calibration, feature extraction, and model training, we demonstrate the feasibility and promising prospects of this method. We have built a database that includes 14 people and 52,805 frames. And our experimental evaluations demonstrate that our algorithm is effective.