← Selected work

COMPUTER VISION · January – May 2026

Athletic Motion Capture System

A three-camera, 75 Hz pipeline that reconstructs athletic movement from synchronized 2D landmarks. Python, OpenCV, MediaPipe, and AprilTag calibration produced 8,256 valid 3D positions for squat analysis through joint angles and skeletal trajectories.

3D skeletal reconstruction viewer
Reconstructed visualization · illustrative pose

Results

75 Hzthree-camera capture
8,256valid reconstructed 3D positions
5.85 pxmean reprojection error

The viewer is a reconstructed visualization using an illustrative pose. Reported results come from the original project, not from this illustration.

Engineering approach

AprilTag-based camera calibration and SVD triangulation reconstruct 3D positions. Calibration succeeded in 475 of 476 frames (99.8%). Analysis of 87 stable squat frames quantified knee/hip angles, five bone lengths, and pelvis/ankle trajectories, identifying three movement events for automated training assessment.

Three-camera landmark views
Three-camera landmark views · Reconstructed illustrative visualization
Explore the other projects ↗