Electromagnetic fields power a wide range of modern technologies, but
human perception is limited to visible light, necessitating alternative systems
for sensing beyond this spectrum. Conventional technologies like cameras,
LiDAR, and high-frequency radar perform well under ideal conditions but
struggle with occlusions, poor lighting, high costs, and privacy concerns.
This paper explores the use of a consumer-grade software-defined radio
(SDR), the LimeSDR Mini 2.0, for robust hand gesture recognition and hand
pose estimation through occlusions such as walls. We introduce two 1D
Convolutional Neural Networks: RF-Gesture, which classifies three distinct
hand gestures with 87% test accuracy across three occlusion levels (clear line of sight, 0.75 cm foam board, and a 4.5 cm wood block), and RF-HandMark, that estimates hand poses using 21 hand landmarks. Leveraging amplitude and phase data from low-frequency (3.4 GHz) and ultra-low bandwidth (500 KHz) Linear Frequency Modulated Continuous Wave Radar (LFMCW), this work demonstrates the potential of affordable SDR-based systems for accurate, low-cost, and privacy-conscious gesture and pose sensing.