Radar Lab Publishes mmPose-FK for Human Pose Estimation via Forward Kinematics
The UA radar group publishes mmPose-FK, integrating mmWave radar with forward kinematics for stable skeletal pose estimation.
Dr. Siyang Cao's lab has published a new method titled mmPose-FK: A Forward Kinematics Approach to Dynamic Skeletal Pose Estimation Using mmWave Radars in IEEE Sensors Journal.
Technical Innovation
This method addresses challenges inherent in mmWave radar sensing — such as low spatial resolution and noise — by incorporating dynamic forward kinematics into the deep learning pipeline. This yields more stable and accurate pose tracking despite measurement uncertainties.
Evolution of mmPose Series
Earlier versions of the lab's pose estimation techniques include:
- mmPose-NLP: Applied natural language processing techniques for skeletal inference
- mmPose: Used CNNs for basic pose estimation
- mmPose-FK: Current state-of-the-art with forward kinematics integration