Highway Visibility Estimation under Adverse Weather Video Dataset

Currently, accurate visibility estimation is crucial for traffic safety and weather forecasting in the environmental meteorological industry. However, the rapid changes in visibility under adverse weather present significant challenges to existing estimation models, with many traditional methods struggling to achieve ideal results due to a lack of high-quality adverse weather data. This dataset aims to provide a rich and diverse video library focusing on changes in highway visibility under various types of adverse weather conditions, supporting the development of more advanced machine learning algorithms. Data is collected using high-resolution camera equipment, covering various weather conditions such as heavy fog, torrential rain, and blizzards, shot in real highway environments. Regarding quality control, the dataset has undergone multiple rounds of annotation and consistency checks, and has been reviewed by an expert team from the field of meteorology and traffic safety, ensuring annotation accuracy of no less than 95%. The annotation team consists of 20 professionals with a background in meteorology. Data preprocessing includes noise filtering, frame deduplication, etc., organized in 'MP4' format storage. Each video segment in the dataset corresponds to weather conditions and visibility labels, facilitating retrieval and analysis. The quality of the dataset lies in its high annotation accuracy and relevance to practical applications, providing over 95% consistency and integrity assurance. Through innovative data augmentation methods like synthetic weather scenes, this dataset excels in addressing hard-to-capture extreme weather conditions, significantly improving the accuracy of visibility estimation models under adverse conditions, with performance improvements near 30% over traditional datasets. As a valuable resource for the environmental and transportation fields, it features high coverage and scarcity of weather scenes, with wide applicability, filling gaps in existing datasets, showing unique advantages when used for training models with strong generalization abilities. Its flexible structure design also supports easy integration of incremental data for future weather conditions and scenes.