Indian Multi-Violation Traffic Dataset (IMVTD)

The Indian Multi-Violation Traffic Dataset (IMVTD) is a real-world dataset designed for detecting multiple traffic violations involving two-wheelers under Indian road conditions. The dataset consists of 6372 images collected in Chennai using mobile cameras, capturing diverse urban traffic scenarios. It includes annotations in YOLO format for multiple object classes related to rider safety, behavior, and vehicle identification. Classes include: two_wheeler, driver, helmet, no_helmet, pillion, license_plate, mobile_phone, and earphone. The dataset is split into training (5817 images), validation (349 images), and test (206 images) sets. Various augmentation techniques such as flipping, rotation, brightness/contrast adjustment, noise injection, and occlusion have been applied to improve model robustness. This dataset is intended for research in traffic monitoring, computer vision, and intelligent transportation systems.