Transfer learning with VGG16 and InceptionV3 for traffic sign classification Available to Purchase

The needs to develop an advance driver assistance have been increasing, especially to lessen the number of accidents by helping drivers control the vehicle. One of the features in ADAS is automatic traffic sign recognition. Recognizing traffic signs can be performed by using machine learning algorithm. The algorithm is used to determine the meaning of the signs through classification algorithms. Several algorithms have been used to perform this task from classical machine learning approach to deep learning. Recent development in deep learning suggests new training method named transfer learning. It essentially uses deep learning architecture that has been trained with ImageNet data set to train on the data set of interest, hence provide better result. Several architectures which perform best in performing image classification in transfer learning are VGG16, ResNet50, Inception, and Xception. This research aims to perform classification task by using transfer learning with VGG16 and Inception architecture. VGG16 is chosen as it is the first architecture that uses transfer learning method. Inception is chosen because despite its smaller size, it is able to perform well. The data set used in this experiment is German Traffic Sign Recognition Benchmark. During experiment, the learning rate, momentum and normalizer is varied. The result of this experiment shows that VGG16 achieves the best accuracy of 99.33%. Meanwhile, the highest accuracy for Inception V3 is 98.44%.