Uncertainty Mixed-Traffic Dataset (UMT-Dataset)

The Uncertainty Mixed-Traffic Dataset (UMT-Dataset) is designed to support the development and evaluation of object detection and tracking algorithms in complex, real-world traffic environments characterized by high uncertainty. Unlike conventional datasets, UMT-Dataset captures nonlinear and non-sequential movement patterns among heterogeneous road users such as cars, motorcycles, pedestrians, trucks, buses, and unstructured vehicles like motorized trishaws. The dataset reflects dynamic interactions, spontaneous direction changes, and bidirectional flows that challenge standard perception systems. UMT-Dataset provides annotated full-HD images in YOLO format, organized into training, validation, and testing subsets, and includes scenarios recorded during both daytime and nighttime. This dataset extends the previously proposed MXT-Dataset by offering greater visual diversity, contextual richness, and a stronger basis for predictive perception modeling in autonomous driving systems.