I’m excited to share the successful completion of my MS (Artificial Intelligence) thesis titled:
🎓 “Intelligent Shot Detection & Cricket Highlights”
📍NED University of Engineering and Technology
👨🏫 Supervised by: Uzair Abid
🔍 Problem Statement:
Cricket highlight generation is traditionally manual, time-consuming, and subjective. At the same time, deeper shot-level analytics for batsmen are rarely automated or visualized effectively.
This research focuses on automating cricket highlight generation while also providing AI-driven batsman shot analysis, using modern Computer Vision and Transformer-based models.
🧠 What I Built (End-to-End AI System)
✅ Event-Driven Highlight Generation
Fine-tuned YOLOv8 models to detect:
Cricket pitch
Score bar
Ball movement
Ball-wise segmentation using pitch & score-bar co-detection
Ball bounce detection via trajectory (Y-axis) analysis
Automatic extraction of highlight-worthy deliveries
✅ Shot Classification using Transformers
Fine-tuned ViViT (Video Vision Transformer) model
Classifies 10 cricket shots, including:
Cover Drive, Pull, Hook, Sweep, Lofted Shot, Straight Drive, etc.
Achieved ~74% test accuracy on custom-labeled datasets
✅ Interactive React-Based Dashboard
Upload full cricket match videos
Real-time processing status
Visual analytics using Bar & Doughnut charts
Ball-wise video previews
Annotated frames + bounce zone visualization
Auto-generated final highlights video
🖥️ System Architecture (How It’s Used)
I also designed a clear SOP (Standard Operating Procedure) so the system can be easily run and tested:
🔹 Backend: Python (Flask) + YOLOv8 + Transformer models
🔹 Frontend: React.js dashboard
🔹 Workflow:
Upload match video
AI processes frames → detects events
Shot classification & bounce analysis
Dashboard displays analytics + highlights
This makes the solution usable not just for research, but for analysts, coaches, and future real-time extensions.
🖥️ System Architecture (How It’s Used)
I also designed a clear SOP (Standard Operating Procedure) so the system can be easily run and tested:
🔹 Backend: Python (Flask) + YOLOv8 + Transformer models
🔹 Frontend: React.js dashboard
🔹 Workflow:
Upload match video
AI processes frames → detects events
Shot classification & bounce analysis
Dashboard displays analytics + highlights
This makes the solution usable not just for research, but for analysts, coaches, and future real-time extensions.
Github Link: https://lnkd.in/dtNPAQf2
I’m proud of how this project bridges AI research with real-world sports analytics.
Always open to feedback, collaboration, and discussions around AI + Computer Vision + Sports Tech 🚀
Top comments (0)