From 26 Keypoints to Clinical Metrics
Two calibrated cameras, DLT triangulation, trajectory filtering, and honest uncertainty intervals: what turns a pose demo into a record a clinician will sign.
Vision in the Real World02
Subject · 7 posts
sports science
Teaching a Camera to Grade Its Own Footage
Camera obliquity is invisible to one lens, so the system infers it from anatomy. Banding its own error by that estimate beats banding by ground truth.
Biomechanics from Video05
Angles Don't Lie: Measuring a Cricket Bowler's Action With a Webcam
Joint angles from pose estimation, event detection inside a 100-millisecond delivery swing, and the frame rate that decides what you are allowed to claim.
Vision in the Real World01
Facing, Sign, and the Error That Mirrors Everything
Which way an athlete faces sets the sign of every sagittal angle at once. One bit, 100 out of 100 on 919 comparisons, and why sign errors are not small.
Biomechanics from Video04
Why 2D Beat 3D on the Angles That Matter
Lifting one camera to 3D sounds strictly better than measuring in the image plane. On sagittal flexion it is worse, and the reason is a free parameter.
Biomechanics from Video03
From Pixels to a Knee Angle
The whole monocular pipeline for one number, and what the detector's pixel error costs once you express it in degrees instead of in average precision.
Biomechanics from Video02
What "Accurate" Means For a Joint Angle
A percentage without a tolerance is not a claim. The full scorecard for a video biomechanics system, by configuration, and why 10 degrees settles it.
Biomechanics from Video01