Andy's Patch Notes

Project! Experimenting with pose detection and CNNs Trying to use computer vision to solve Judo!
Built a real-time judo throw recognition system using a CNN-based pose detection pipeline running at 60 FPS. Extracted kinematic keypoint sequences from video frames and trained a classifier to distinguish between throw types based on joint trajectory patterns. Achieved 63% classification accuracy across multiple throw categories, the main bottleneck was feature representation, since similar throws share a lot of overlapping keypoint motion early in the sequence.
Shown here is a Koga Ippon Seoi-Nage