Adverse Weather Dataset

This dataset contains motion features and appearance cues extracted from video sequences, specifically designed for early adverse weather classification, detection tasks. The motion features are obtained from optical flow color maps, while the appearance cues are extracted from images of heavy fog, light fog, smoke, snow and clear weather. The dataset is categorized into five classes: heavy fog, light fog, smoke, snow and clear weather. Each class contains two types of data components: original images and optical flow color maps. In total, the dataset includes 1,990 images, evenly split between the five classes, 389 heavy fog, 401 light fog, 341 smoke, 440 snow and 419 clear weather images. Additionally, each class includes same number of optical flow color maps. The optical flow color maps are generated using a fractional order variational model. To effectively capture both static (appearance-based) and dynamic (motion-based) features. These fused representations serve as input for training deep learning models for early adverse weather detection.