Joint estimations and corrected errors for Robust Skeletal Motion Tracking, samples

This study uses a custom dataset of upper-limb rehabilitation exercises recorded with two synchronized cameras, capturing both frontal and lateral perspectives. Each exercise session involves participants performing standard therapeutic movements (e.g., arm flexion, biceps curl) while wearing various clothing types. The dataset contains:

Excel Data of Marker Coordinates:

Files listing all tracked keypoints (markers)—including elbow, wrist, shoulder, etc.—for each exercise.

Mean and standard deviation of raw depth error, alongside the corrected depth values computed by dual-camera pipeline.

Summaries of specific markers of interest (e.g., right wrist, right elbow) used for further analyses.

The videos are not publicly distributed due to privacy requirements. However, researchers may obtain access to these recordings, with the reasonable request to authors.

The dataset was assembled under a data-protection protocol to ensure confidentiality and to comply with institutional review board regulations. As such, the openly released component is limited to the Excel sheets with marker coordinates and depth adjustments, while raw video data remains restricted.