Traffic Sign Detection and Recognition using Deep Learning

Abstract

Detecting and recognizing traffic signs automatically is vital for efficiently managing traffic-sign inventory with minimal human intervention. While existing methods in the computer vision field excel at recognizing signs pertinent to advanced driver-assistance and autonomous systems, they cover only a fraction of the total traffic sign categories, leaving a significant portion unaddressed. This gap poses a challenge for automating traffic-sign inventory management, which requires handling a broader range of signs. In our study, we tackle this challenge by employing a convolutional neural network (CNN) approach, specifically the mask R-CNN, to handle the entire detection and recognition process in a seamless manner. We propose several enhancements to improve the detection performance, which we evaluate on a dataset comprising 200 different traffic sign categories, including those not previously explored. These findings indicate the feasibility of deploying our approach in real-world applications for traffic-sign inventory management.