This is the official repository for the SIVD dataset, which contains Iranian vehicle images for real-time multi-camera video tracking and recognition. The dataset and trained models are publicly available and can be downloaded from Google Drive.
In this paper, a new publicly available web-Scraped Iranian Vehicle Dataset (SIVD) for simultaneous real-time vehicle tracking and recognition is proposed. The datasets provided for Iranian cars in the literature have two fundamental problems. First, the lack of images from different angles, and second, the small number of classes compared to the dispersion of car models in the real world. Therefore, for the purposes of this paper, Iranian vehicle images from car sales websites are collected, and the SIVD dataset is proposed which contains 29 classes and 36,705 images. This paper aims at developing a classification network for Iranian vehicle recognition and implement a real-time tracking system using the YOLOv5 network to perform real-time vehicle model recognition and tracking tasks simultaneously. The ResNet50 achieved an accuracy of 99.29%, the highest among the investigated classification networks.
The SIVD dataset is described in detail in our paper, SIVD: Dataset of Iranian Vehicles for Real-Time Multi-Camera Video Tracking and Recognition, which was presented at the 2022 8th Iranian Conference on Signal Processing and Intelligent Systems (ICSPIS). If you use this dataset in your research, please cite our work using the following BibTeX entry:
@INPROCEEDINGS{siahkali2022sivd,
author={Siahkali, Farbod and Alavi, Seyed Amirmahdi and Masouleh, Mehdi Tale},
booktitle={2022 8th Iranian Conference on Signal Processing and Intelligent Systems (ICSPIS)},
title={SIVD: Dataset of Iranian Vehicles for Real-Time Multi-Camera Video Tracking and Recognition},
year={2022},
volume={},
number={},
pages={1-7},
doi={10.1109/ICSPIS56952.2022.10043932}}
Our research was conducted in the Human and Robot Interaction Laboratory (TaarLab) at the University of Tehran. If you have any questions or comments, please feel free to contact us at siahkali@purdue.edu or farbodsiahkali80@ut.ac.ir.