Abstract: Traffic signs are essential in regulating traffic on the road, guiding drivers, thus helping to avoid injuries, property damage, and deaths. Managing traffic signs with automatic detection and recognition is a significant component of any Intelligent Transportation System (ITS). In the age of self-driving vehicles, the importance of automatic detection and recognition of traffic signs cannot be emphasized enough. This paper introduces a deep-learning-driven autonomous approach for the identification of traffic signs in India. The automatic detection and recognition of traffic signs were designed using a Convolutional Neural Network (CNN)- Refined Mask R-CNN (RM R-CNN)-based end-to-end learning framework. The proposed concept was evaluated using an innovative dataset featuring 6480 images that included 7056 instances of Indian traffic signs categorized into 87 classes. We provide multiple enhancements to the Mask R-CNN model in terms of both architecture and data augmentation. We introduce multiple improvements to the Mask R-CNN model, both in terms of architecture and data augmentation. We have examined particularly difficult Indian traffic sign categories that have not been documented in earlier research. The dataset for training and testing our proposed model is gathered by taking images in real-time on Indian roads. The evaluation findings show an error rate of less than 3%. Additionally, the performance of RM R-CNN was contrasted with traditional deep neural network architectures like Fast R-CNN and Mask R-CNN. Our proposed model attained a precision of 97. 08%, which surpasses the precision achieved by the Mask R-CNN and Faster R-CNN models.
Keywords: Traffic Signs, Intelligent Transportation System (ITS), Refined Mask R-CNN (RM R-CNN), Indian Traffic Signs Dataset, Deep Learning, Real-time Image Processing, Data Augmentation, Precision Rate.
Copyright Form