Transforming Basketball Analytics with AI, Deep Learning, and Computer Vision — Real-Time Detection, Tracking, and Insights.
Explore Project 🚀This project showcases a real-time AI-powered system designed to automate basketball gameplay analysis. It detects players, ball, and rim, tracks movements, classifies teams, and annotates gameplay actions efficiently using Deep Learning, UMAP, KMeans, and advanced visualization techniques.
GPU-accelerated frame extraction and stride sampling with ONNX Runtime & Supervision.
Custom-trained YOLOv8 and Roboflow models for detecting players, ball, and rim.
ByteTrack for multi-object tracking and UMAP + KMeans for team classification.
| Metric | 10 Epochs | 15 Epochs |
|---|---|---|
| mAP50 | 93.4% | 94.2% |
| mAP50-95 | 71.1% | 73.7% |
| Precision | 85.7% | 89.4% |
| Recall | 90.1% | 91.0% |