Review of Car Make & Model Recognition Systems

This paper focuses attention towards the review of various applications and approaches in the field on image processing up to and including recent advancement of deep learning using convolutional neural networks that can be used as tools for tackling the obstacles of Car Make and Model Recognition (CMMR) in real-world environment images. Such algorithms for CMMR system are typically designed to detect specific features in images that used to be formed by feature engineering processes and are now being replaced with deep learning. The review consists of three types of algorithms. The first set explores the traditional methods that use feature extraction to localise cars in various applications and attempt to provide solution for recognizing car characteristics with feature matching over whole images in database. The next set under consideration was deep learning since it demonstrated promising results due to automatic feature engineering although still being an area under consistent research and improvement over the past few years. This paper refers to how the deep learning systems have contributed towards successful CMMR and not a comparison of deep learning architectures. The last section of this review is focused in fine-grained classification with deep learning. This is conducted especially considering the cars that are generally built up of many different parts and identifying them based on fine-grained parts from very recent researches and whether it is a viable method for attaining better overall classification accuracy score.