RT-OCR : Real-Time OCR Pipeline with Formal Timing Guarantees for Edge Vision Systems

Deep learning OCR achieves high accuracy but exhibits unpredictable latency, preventing deployment in safety-critical systems like autonomous vehicles and industrial robotics. RT-OCR provides formal worst-case execution time (WCET) bounds on edge platforms through a timing-aware pipeline architecture, adaptive workload controller, and measurement-based WCET characterization on Jetson Nano and Raspberry Pi 5. Evalua- tion demonstrates 73% jitter reduction, 32% WCET improvement, and 97.5% accuracy with 100% deadline compliance across 10,000 consecutive frames.