Driving into the Future: Artificial Intelligence based Traffic Sign Recognition using Learning Assisted OCR Principle

Abstract

Managing the inventory of traffic signs relies heavily on automatic detection and recognition. It offers a precise and quick method of managing traffic-sign inventory with little human intervention. The detection and identification of traffic signs is an issue that has received a lot of attention from the computer vision field. The traffic signs required for sophisticated driver-assistance and self-driving vehicles are well-served by most current methods. But it’s just a fraction of all roadway markings (approximate 40-50 kinds among hundreds), and how well it will work with the rest of the traffic signs—the ones that need to be automated so that traffic-sign inventory management doesn’t require any human intervention—is still up in the air. The goal of this work is to automate the process of managing traffic signs’ inventory by addressing the challenge of detecting and recognizing a large number of sign classifications. This study introduces a new approach called Learning Assisted OCR Principle (LAOCRP). To assess its performance, it is cross-validated with the traditional model called Generic OCR (GOCR). The outcomes are shared for extremely difficult traffic-sign classes that have not been addressed in prior research. Our thorough evaluation of the deep learning method for traffic sign detection with high intra-category appearance variation yields lower error rates in comparison to the state-of-the-art methods, making them suitable for use in real-world traffic sign inventory management applications.