Other pose systems compared

Markerless systems differ in what they track, how many cameras they need, where the computation runs and how much effort they take. Marker-based laboratory systems remain the reference for validation.

Overview

System	What	Cameras	Runs	Note
SInES Pose Tracker (BlazePose)	33 body points, 2D and model-based 3D	one	browser, local	no installation; 3D per camera only
MediaPipe (Google)	body, hands, face (the models behind BlazePose)	one	Python, local	for batch processing and own scripts
VioPose	violin-specific 4D pose with sound	one or more	Python, local	research code, installation effort
OpenPose (Carnegie Mellon University)	body, feet, hands, face; many persons	one or more	C++/Python, local, GPU	long-established; heavy installation; check the licence
OpenCap (Stanford)	3D kinematics and estimated joint loads	two or more phones	cloud	videos are uploaded – data protection and consent
FreeMoCap	3D motion capture	several webcams/phones, calibrated	local, open source	free; calibration needed
SPD method	body, hands, instrument and bow in 3D	many	research pipeline	shows what many views and sound add
Qualisys, Vicon (markers); Noraxon (inertial sensors, EMG)	reference movement data, muscle activity	lab system	laboratory	for validating the markerless measures

OpenPose: use the official source

The official OpenPose is the project of the Perceptual Computing Lab at Carnegie Mellon University: github.com/CMU-Perceptual-Computing-Lab/openpose. Websites with similar names that offer an “OpenPose” download are not connected to it – do not install software from them.

Choosing

Teaching, quick studies, privacy: SInES in the browser.
Many files, own analysis scripts: MediaPipe in Python, then PyEyesWeb or R.
Violin-specific detail: VioPose, compared with SInES on the same videos.
Accuracy questions: a session with a marker-based system or several calibrated cameras in parallel.

Literature

Cao, Z., Hidalgo, G., Simon, T., Wei, S.-E., & Sheikh, Y. (2018). OpenPose: Realtime multi-person 2D pose estimation using part affinity fields [Preprint]. arXiv. https://doi.org/10.48550/arXiv.1812.08008
Uhlrich, S. D., Falisse, A., Kidziński, Ł., Muccini, J., Ko, M., Chaudhari, A. S., Hicks, J. L., & Delp, S. L. (2023). OpenCap: Human movement dynamics from smartphone videos. PLOS Computational Biology, 19(10), e1011462. https://doi.org/10.1371/journal.pcbi.1011462
Bazarevsky, V., Grishchenko, I., Raveendran, K., Zhu, T., Zhang, F., & Grundmann, M. (2020). BlazePose: On-device real-time body pose tracking [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2006.10204