The skeleton fusion tool reads two frames from two cameras and gets two skeletons from the OpenPose [1], and the Graduate Non-Convexity [2] optimization algorithm is implemented to obtain a single robust 3D skeleton.
Cuda (tested 10.2, 11.8)
*cudnn (tested 7.6.5, 8x)
OpenPose [1] (1.7.0)
Apriltag [3]
[1] OpenPose (https://github.com/CMU-Perceptual-Computing-Lab/openpose)
[2] Graduate Non-Convexity from the work: Robust Event Detection based on Spatio-Temporal Latent Action Unit using Skeletal Information (https://ieeexplore.ieee.org/abstract/document/9636553)
[3] Apriltag (https://april.eecs.umich.edu/software/apriltag)