Real-Time Esports Camera Switching OCR System
A high-performance computer vision system for esports tournaments, enabling automated camera switching and real-time OCR-based data extraction.
Certificate
Project Overview
Key Achievements
- Developed real-time camera switching software for esports tournaments, enhancing live streaming quality and viewer experience.
- Built a high-speed OCR system capable of extracting structured data in real time (120ms per image) with millisecond-level accuracy.
- Engineered and optimized a computer vision pipeline by experimenting with Tesseract, YOLO, PaddleOCR, Qwen, LLaMA, and other advanced AI models.
- Successfully delivered a complex project that no industry competitor had achieved at the time.
Technologies Used
Project Details
I have worked on this computer vision system and have experimented with many thing to make the work in the most efficient way.
The core challenge was to create an automated camera switching system during live tournaments, extracting game data through OCR (Optical Character Recognition) technology.
I have experimented with various methods including Tesseract, YOLO, PaddleOCR, Qwen, Llama and lot more. I was able to optimize the pipeline to achieve optimal performance getting output in 120ms with high accuracy. This was essential for live broadcasting where delays could significantly impact viewer experience.
This project was complex, although many people have tried to do this, no one has yet achieved it. I had worked a lot on this project and was glad to deliver it.