JUIVCDv1: Development of a still-image based Dataset for Indian Vehicle Classification

Abstract

Designing an automatic vehicle classification system from still images or videos would be highly beneficial for developing a traffic control system. On automatic vehicle classification, numerous articles have been published in the literature. Over the years, researchers in this subject have created and used a variety of databases, but most often, these databases are not found to be appropriate in Indian scenarios due to the specific peculiarities of the road conditions, nature of congestion, and vehicle types usually seen in India. This thesis primary goal is to create a new still image database called the JUIVCDv1 that contains 12 different vehicle classes that were gathered utilising mobile phone cameras in a variety of ways for developing an automated traffic management system. We have also mentioned the characteristics of the current databases and the various factors we took into account when creating the database for the Indian scenario. Apart from this, we have benchmarked the results on this database using a five-base model architecture. Five base models are used: Efficient- Net, InceptionV3, DenseNet121, MobileNetV2, and VGG19. Among these five-base models, EfficientNet achieved the best accuracy, i.e., 93.82%.