International Research Journal of Education and Technology

This paper presents a real-time video text tracking system integrating deep learning with IoT-enabled edge processing for efficient text detection, tracking, and recognition. Using the CRAFT text detector, DeepSORT for tracking, and Tesseract OCR for recognition, the system ensures low-latency performance with AWS EC2 and NVIDIA A100 GPU acceleration, while IoT devices enable edge computing to reduce bandwidth usage. Docker containerization ensures consistency, PostgreSQL manages structured text data, and a FastAPI-based API enables real-time text retrieval. Evaluations on ICDAR and TextVQA datasets demonstrate resilience to motion blur, occlusions, and varying text orientations, making it suitable for applications like automated transcription, surveillance, and smart retail. Future work will focus on enhancing adaptive recognition, optimizing IoT-edge processing, and improving scalability.

Keywords— Video text tracking, real-time text recognition, optical character recognition (OCR), text detection, machine learning, vehicle identification, CCTV analysis.