Current Datasets and Their Inherent Challenges for Automatic Vehicle Classification

Automatic Vehicle Classification (AVC) systems have become a need of the hour to manage the ever-increasing number of vehicles on roads and thus maintain a well-organized traffic system. Researchers around the world have proposed several techniques in the last two decades to address this challenge. However, these techniques should be implemented on realistic datasets to evaluate their efficiency in practical situations. Hence, it is understood that for the success of this domain, datasets play an important role, mostly publicly accessible by the research community. This article presents a comprehensive survey regarding various datasets available for solving AVC problems such as vehicle make and model recognition (VMMR), automatic license plate recognition, and vehicle category identification during the last decade. The datasets are categorized into two types: still image-based, and video-based. Again, the still image-based datasets are further classified as aerial imagery-based and front image-based datasets. This study has presented a thorough comparison of the different types of datasets with special reference to their characteristics. This study also provides an elaborative analysis of all the datasets and suggests a few fundamental future research scopes toward AVC. This survey can act as a preliminary guideline for researchers to develop a robust AVC system specially designed as per their needs and also to choose suitable datasets for comparing their models.