Traffic Sign Recognition System using Ensemble based Deep Learning Models

With the rapid growth of movement in automobiles, the automatic traffic sign recognition (TSR) becomes an important area of research in recent time. Various deep learning techniques are proposed for developing TSR system using supervised frame-work in recent past. Many of the existing individual supervised models may not produce required accuracy due to scarcity of the sufficient training images. In this context, ensemble learning may be a viable alternative by assimilating the decisions of the individual classifiers which aims to achieve better accuracy. This article presents an efficient ensemble based deep learning method amalgamating the decisions of CNN, ResNet50 and GoogLeNet for automated TSR system applied on the GTSRB (German Traffic Sign Recognition Benchmark) dataset. The experimental results demonstrate better classification performance produced by proposed ensemble based deep neural network than individual constituent deep learning classifiers and six other existing counterpart methods.