Research highlights
Antoine Falisse et al., developed a more accurate and generalizable model, named "marker enhancer", to predict the position of 43 anatomical markers from 20 keypoints identified from video. We trained this model on a large database of 1,433 hours of data from 1,176 subjects.
This article shows this model improves kinematic accuracy (4.1° error) compared to OpenCap's original model (Uhlrich S & Falisse A, et al. (2023)) which had 5.3° error on a benchmark dataset. We also showed that it better generalized to unseen, diverse movements (4.1° error) than OpenCap’s original model (40.4° error).
Links to original paper
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10844513
