chonyy/AI-basketball-analysis

Leverages OpenPose for real-time human pose estimation to extract body keypoints (elbow, knee angles) during shots, combined with Faster R-CNN object detection trained on COCO dataset to classify shot outcomes. Provides both a Flask web interface for video/image uploads and a REST API endpoint (`/detection_json`) returning JSON-formatted keypoint coordinates and detection confidence scores. Requires GPU with CUDA support for efficient processing, though Google Colab integration enables GPU-free experimentation.

1,224 stars. No commits in the last 6 months.