OpenPose

The first real-time multi-person system to jointly detect human body, hand, facial, and foot keypoints on single images.

Developed by the CMU Perceptual Computing Lab, OpenPose utilizes a bottom-up approach via Part Affinity Fields (PAFs) to associate body parts with specific individuals. It tracks 135 total keypoints (body, hands, face, and feet) and maintains high performance regardless of the number of people in the frame. This library is the industry standard for markerless motion capture, powering applications in sports analytics, VR/AR, and clinical gait analysis. It supports Caffe, TensorFlow, and PyTorch backends while running efficiently on NVIDIA GPUs via CUDA and cuDNN.

https://github.com/CMU-Perceptual-Computing-Lab/openpose