As part of the V2 Gnocchi update, OpenPTrack now uses machine learning for pose recognition alongside person tracking and its new object tracking capabilities. OPT pose recognition extends the OpenPose skeletal tracking library to multiple cameras, and includes the ability to train the system to detect unique poses.
Early adopters include UCLA and Indiana University STEP researchers, who are integrating pose recognition into their work with elementary school students. As they learn about scientific phenomenon, and their positions control forces within nature, the students can now also interact with these forces by signaling and with other gestures.
Documentation on how to use and implement pose recognition and its associated training methods is forthcoming, and will be included in the OpenPTrack V2 wiki. More information the OPT pose recognition work can be found in this paper by the OPT team.
UPDATE: Read our new overview of how 3D skeleton tracking and pose recognition work with OPT.