UNPOSED: an Ultra-wideband Network for Pose Estimation with Deep Learning

Human pose estimation refers to the detection and tracking of the position of human body parts from input data like images and videos. The most common methods are based on computer vision techniques that require the installation of cameras such as RGB or infrared cameras. However, vision-based systems are susceptible to challenges like complicated backgrounds, occlusions, and low-resolution images. Recently, UltraWide-Band (UWB) technology has gained popularity due to its low complexity, high accuracy, and penetration capability, hence we propose a novel approach for human pose estimation using UWB technology. The solution retrieves the 3D human pose using five body-mounted UWB sensors placed at the pelvis and at limbs joints, namely wrists and ankles. We designed a transformer-based architecture capable of reconstructing the human skeleton from a small subset of joints measured with ranging information using UWB.