Document Type : Article
Faculty of Electrical Engineering, Shahrood University of Technology, Shahrood, Iran
School of Electrical Engineering, Iran University of Science and Technology, Tehran, Iran.
Department of Computer Engineering, Ferdowsi University of Mashhad, Mashhad, Iran
Fine-grained vehicle type recognition using on-road cameras is among interesting topics in machine vision. It has several challenges like inter-class similarity, different viewing angles, and different lighting and weather conditions. This paper presents a novel approach for vehicle classification based on a novel augmentation method and deep learning. In the proposed smart augmentation, the vehicle images of each class are registered on the reference vehicles of all other classes and then added to the training set of that class. In this way, we will have a lot of new images which are very similar to both reference and target classes. This helps the CNN model to handle inter-class similarities very well. In the test phase, the input image is registered on every reference image in parallel and applied to the model. Finally, the winner is determined by summing up the provided scores of all models. The targeted data augmentation along with the proposed classification strategy has high recognition power and is capable of providing high accuracy using small CNNs or any other classification method without the need for large datasets. The proposed method achieved a recognition rate of 99.8% with only 150K parameters.