A Detailed Study on Bangladeshi Road Sign Detection and Recognition

Abstract

Automatic road sign detection and recognition is crucial for autonomous Driver Assistance Systems (DAS). Numerous road accidents and casualties are reported every year due to the negligence of drivers. The system is mainly developed for autonomous vehicles and drives to warn them about the existing road sign on the road and also to reduce the frequency of accidents throughout the country. Road sign detection and recognition is still challenging concerning Bangladesh because of the shortage of big dataset. The dataset for this work has been collected from different regions throughout the country. In this system, the Single Shot Multibox Detector (SSD) is used for automatic detection and recognition of road sign. The Convolutional Neural Network (CNN) is also applied for the recognition purpose. The system is able to achieve the average detection rate of 76.52% and the recognition accuracy of 86.23% using SSD and also accomplish the recognition accuracy of 80.26% while using CNN. Therefore, the proposed system is robust enough to detect and recognize road sign.