Marker-based motion capture (mocap) is a conventional method used in biomechanics research to precisely analyze human movement. However, the time-consuming marker placement process and extensive post-processing limit its wider adoption. Therefore, markerless mocap systems that use deep learning to estimate 2D keypoint from images have emerged as a promising alternative, but annotation errors in training datasets used by deep learning models can affect estimation accuracy. To improve the precision of 2D keypoint annotation, we present a method that uses anatomical landmarks based on marker-based mocap. Specifically, we use multiple RGB cameras synchronized and calibrated with a marker-based mocap system to create a high-quality dataset (RRIS40) of images annotated with surface anatomical landmarks. A deep neural network is then trained to estimate these 2D anatomical landmarks and a ray-distance-based triangulation is used to calculate the 3D marker positions. We conducted extensive evaluations on our RRIS40 test set, which consists of 10 subjects performing various movements. Compared against a marker-based system, our method achieves a mean Euclidean error of 13.23 mm in 3D marker position, which is comparable to the precision of marker placement itself. By learning directly to predict anatomical keypoints from images, our method outperforms OpenCap's augmentation of 3D anatomical landmarks from triangulated wild keypoints. This highlights the potential of facilitating wider integration of markerless mocap into biomechanics research. The RRIS40 test set is made publicly available for research purposes at koonyook.github.io/rris40 .

Insightful feedback generation for daily home-based stroke rehabilitation is currently unavailable due to the inefficiency of exercise inspection done by therapists. We aim to produce a compact anomaly representation that allows a therapist to pay attention to only a few specific sections in a long exercise session record and boost their efficiency in feedback generation. This study proposes a data-driven technique to model a repetitive exercise using unsupervised phase learning on an artificial neural network and statistical learning on principal component analysis (PCA). After a model is built on a set of normal healthy movements, the model can be used to extract a sequence of anomaly scores from a movement of the same prescription. The method not only works on a standard marker-based motion capture system but also performs well on a more compact and affordable motion capture system based-on Kinect V2 and wrist-worn inertial measurement units that can be used at home. An evaluation of four different exercises shows its potential in separating anomalous movements from normal ones with an average area under the curve (AUC) of 0.9872 even on the compact motion capture system. The proposed processing technique has the potential to help clinicians in providing high-quality feedback for telerehabilitation in a more scalable way.

This paper presents a calibration approach for multiple synchronized global-shutter RGB cameras surrounding a large capture volume for 3D application. The calibration approach uses an active wand with two LED-embedded markers waved manually within the target capture volume. Data from the waving wand are combined with chessboard images taken at close range during each camera’s intrinsic calibration, optimizing camera parameters via our proposed bundle adjustment method. These additional constraints from the chessboard are developed to overcome an overfitting issue of wand-based calibration discovered by benchmarking its 3D triangulation accuracy in an independent record against a ground-truth trajectory and not on the record used for calibration itself. Addressing this overfitting issue in bundle adjustment leads to significant improvements in both 3D accuracy and result consistency. As a by-product of this development, a new benchmarking workflow and our calibration dataset that reflects realistic 3D accuracy are proposed and made publicly available to allow for fair comparisons of various calibration methods in the future. Additionally, our experiment highlights a significant benefit of a ray distance-based (RDB) triangulation formula over the popular direct linear transformation (DLT) method.