Foggy License Plates Worldwide: A Comprehensive Dataset

The "Foggy License Plates Worldwide" dataset is a specialized collection designed to advance the recognition and detection of vehicles and license plates under foggy conditions. This dataset includes 4420 2D-RGB images, featuring a diverse array of vehicles from Bangladesh, Thailand, Saudi Arabia, and other regions with English license plates. These images are derived from a secondary dataset and have been augmented using monocular depth estimation to simulate varying degrees of fog, offering a realistic set of data for challenging visibility scenarios.

The dataset is not limited to one type of vehicle but includes buses, trucks, CNG vehicles, motorcycles, cars, and standalone license plates, providing a broad spectrum for analysis. While the Bangladeshi subset contains 2754 annotated images, the dataset also includes 388 images from Thailand, 433 from regions with English license plates, and 845 from Saudi Arabia, though the latter does not come with annotations.

This variety is crucial for developing robust algorithms that can operate across different regions and vehicle types, especially in foggy conditions. The use of the Monodepth2 Network to artificially introduce fog effects based on depth estimation ensures that the dataset can mimic real-world scenarios, enhancing the development of automated systems for license plate recognition, vehicle detection, and traffic monitoring under adverse weather conditions. By offering a global perspective with plates from multiple countries, this dataset serves as an invaluable resource for researchers and developers in the field of computer vision, aiming to enhance the accuracy and reliability of systems in foggy environments.

“全球雾天车牌”（Foggy License Plates Worldwide）数据集是一款专为推进雾天环境下车辆与车牌识别检测技术发展而打造的专业数据集。该数据集包含4420张2D-RGB图像，涵盖了来自孟加拉国、泰国、沙特阿拉伯及其他使用英语车牌地区的多种车型。这些图像源自二级数据集，并通过单目深度估计技术进行数据增强以模拟不同浓度的雾效，可为低能见度挑战性场景提供逼真的测试数据集。

该数据集并未局限于单一车型，涵盖了巴士、货车、压缩天然气（CNG）车辆、摩托车、轿车以及独立车牌样本，可为研究分析提供广泛的覆盖范围。其中孟加拉国子集包含2754张带标注图像，此外数据集还包含来自泰国的388张图像、使用英语车牌地区的433张图像，以及来自沙特阿拉伯的845张图像（后者未附带标注）。

这种多样化的样本对于开发可适配不同地区与车型的鲁棒算法至关重要，尤其是在雾天环境下。研究团队通过Monodepth2网络基于深度估计人工生成雾效，确保数据集能够逼真还原真实场景，从而推动恶劣天气下车牌识别、车辆检测及交通监控自动化系统的研发。该数据集涵盖多个国家的车牌样本，具备全球化视角，可为计算机视觉领域的研究者与开发者提供宝贵的资源，助力提升雾天环境下各类系统的识别精度与可靠性。