Real-Time Snooker Ball Detection System
A fully cloud-deployed AI vision system that detects and tracks all 22 snooker ball classes in real time from a mobile camera - GPU inference via WebSocket, automatic session recording, and browser-accessible from any phone without an app install.
Key Outcome
22-class real-time detection
Live GPU-backed snooker ball tracking from any phone browser - zero hardware beyond a camera.
Snooker coaching and game analysis relied entirely on manual observation. No affordable tool could identify individual ball positions and types in real time from a mobile camera, work overhead from a phone, or automatically record sessions for post-match review.
Trained a YOLOv8m model on a custom dataset covering 22 snooker ball classes (all colours, cue ball, pocket states). Built a FastAPI WebSocket server that receives live camera frames, runs GPU inference on a Tesla T4 on GCP, and streams annotated video back in real time. Mobile-first UI with HTTPS for browser camera access, and automatic MP4 session recording with a review gallery.
Key Features
YOLOv8m trained on 22 snooker ball classes at 1280px resolution
Real-time WebSocket streaming - annotated frame returned per capture cycle
NVIDIA Tesla T4 GPU inference on Google Cloud Platform for sub-second latency
Mobile-first UI with rear camera access via HTTPS (no app install required)
Automatic MP4 session recording saved to gallery for post-match review
GCP Compute Engine deployment with systemd auto-restart on reboot
mAP@50 optimised on a purpose-built snooker detection dataset
Custom confidence threshold controls for different table and lighting conditions
Technology Stack
Project Visuals
Interested in a similar project?
Tell us your requirements and we will respond within one business day.