| Summary: | Vehicle make and model recognition plays an important role in monitoring traffic in a vehicle surveillance system. Identifying vehicle make and model is a challenging task due to intraclass variation, view-point variation, and different illumination conditions (Hassan et al., 2021). In this domain, many datasets regarding car make and model e.g. Stanford Car (Krause et al., 2013), VMMRdB (Tafazzoli et al., 2017, Yang et al., 2015), have already been experimented with by different researchers. However, most of the images in these datasets are high-quality images with no illumination conditions. Further, these images are collected through web crawling or image scraping. This enabled the researchers to achieve good results using deep learning models (Luo et al., 2015). In this article, we have presented an image dataset of 3847 images, designed from high-resolution (1920 1080) videos collected from camera units installed on a highway at different viewpoints with variable frame rates. This helped in collecting images demonstrating a real-world scenario and made this dataset more challenging. Due to consideration of different viewpoints and illumination effects, the dataset will help researchers to evaluate their machine learning models on realworld data (Manzoor et al., 2019).
|
|---|