VGG16 & DenseNet121 based Ensemble Deep Learning Approach for Traffic Sign Recognition
One of the most important features of Advanced Driver Assistance Systems (ADAS) and Intelligent Transportation Systems (ITS), namely Traffic Sign Recognition (TSR), is a computer vision system that can automatically recognize and classify traffic signs to ensure road safety. In the proposed research work, a dataset provided by Kaggle on traffic signs containing 6117 images of various categories of traffic signs such as speed limit, stop, no entry, yield, pedestrian crossing, etc., has been used to propose a framework for classification of traffic signs using Deep Learning (DL) techniques. In the proposed research work, a combination of two architectures, namely VGG16 and DenseNet121, has been used by employing feature level ensemble techniques to improve the efficiency of the developed ensemble architecture by comparing the efficiency of the proposed ensemble architecture with existing architectures. The developed ensemble architecture has been found to attain an accuracy of 99.54%, precision of 99.69%, recall of 99.54%, and F1-score of 99.57%, which shows that the proposed ensemble architecture is productive and reliable compared to existing architectures.