Enhancing Traffic Sign Detection Using a Hybrid Model of DenseNet121 and InceptionV3 with Optimized Dense Layers
Rahul Kumar Gupta, Binayak Ojha, Samir Kumar Majhi, Debendra Muduli, Suman Bashyal, Abhinav Kumar Singh, Anish Ghimire · 2024
Traffic Sign Detection and Recognition models are essential for enhancing the safety and efficiency of autonomous vehicles. This paper presents a novel deep learning approach that with the combination of DenseNet121 and InceptionV3 architectures through a fusion model. Utilizing transfer learning, we adapt these pre-trained convolutional neural networks for these kinds of sign recognition in the Indian context. To enhance model performance, we implement data augmentation techniques during preprocessing, significantly increasing the diversity of the training dataset. Our experiments on the ICTS (Indian Cautionary Traffic Sign) dataset yield impressive results, achieving a classification accuracy of 99.49%, surpassing traditional methods. This work demonstrates the potential of integrating advanced CNN architectures to improve the dependability and precision of traffic sign detection systems, setting the stage for smarter autonomous driving solutions.