Overview

Ball View is a real-time cricket analysis system that captures live video frames from a browser, processes them through a YOLO-based object detection backend, and overlays trajectory predictions and contextual match data back onto the video stream.

---

System Architecture

The system operates as a closed feedback loop across three layers:

1. Browser Layer — A Chrome extension / userscript captures frames from a live HTML5 video player at ~30fps and streams them to the backend via WebSocket.
2. Processing Backend — A FastAPI Python server receives frames, runs YOLOv8 object detection (ball and player tracking), estimates ball trajectories, and performs OCR on the scoreboard to extract match context.
3. Data Sync & Overlay — OCR output is fuzzy-matched against Cricsheet ball-by-ball data to identify the current bowler, batter, and delivery context. Trajectory predictions and detection results are sent back to the browser for canvas overlay rendering.

---

Key Capabilities

• Object Detection — YOLOv8 identifies ball and player positions in each frame, outputting bounding boxes with confidence scores
• Trajectory Estimation — Sequences of ball center coordinates are projected into 3D to predict ball path, pitch map, and speed (kph)
• OCR Scoreboard Reader — Extracts live score and over information to sync with historical ball-by-ball data
• Real-Time Feedback — JSON payloads with object positions, trajectory paths, predicted delivery outcomes (e.g., LBW), and contextual metadata are streamed back to the browser at frame rate

---

Tech Stack

• Backend: Python, FastAPI, YOLOv8, OpenCV, NumPy
• Frontend: JavaScript (Chrome Extension + Userscript), HTML5 Canvas
• Transport: WebSocket (binary frames + JSON feedback)
• Data: Cricsheet match data (JSON/CSV)