Abstract: The automatic detection and recognition of traffic signs is crucial for managing the inventory of traffic signs. With the least amount of human effort, it offers an accurate and timely manner of monitoring traffic-sign inventory. In the realm of computer vision, the recognition and detection of traffic signals is a well-researched topic. In-depth driver-assistance and autonomous systems successfully use the vast majority of current technologies to understand traffic signs. It is yet unknown how well the remaining traffic signs will perform when used to replace manual labor-intensive traffic-sign inventory management because this only covers a small part of all traffic signs (about 50 categories out of several hundred). This study focuses on the difficulty of locating and categorising a large variety of categories for traffic signs that may be used to inventory management software. We employ a convolutional neural network (CNN) method dubbed the mask R-CNN to address the whole pipeline of detection and identification in autonomous end-to-end learning. We offer a number of suggestions that improve overall performance and are evaluated on the ability to recognise traffic signs. Using this technique, 200 traffic-sign types represented in our novel dataset are found. The results are reported for very challenging traffic-sign categories that haven't yet been included in prior studies. We provide a comprehensive analysis of the deep learning method for identifying traffic signs with significant intra-category appearance variation and show below 3% mistake rates.
Keywords: Traffic sign recognition, traffic sign detection, image processing, convolutional neural network
|
DOI:
10.17148/IJARCCE.2023.12255
