Brazilian Urban Traffic Dataset Extended

This dataset comprises 10,000 images collected from real-time public traffic camera video streams made available online by municipal governments in three Brazilian cities: São Paulo (SP), Rio de Janeiro (RJ), and Manaus (AM). These cities were selected because they rank among the seven largest cities in Brazil and provide open streaming infrastructure for traffic monitoring. The dataset targets multi-class vehicle detection in real-world urban traffic scenes, annotated with bounding boxes across four object classes: car, vehicle, motorcycle, and bus. Public object detection benchmarks for traffic scenes are often dominated by data from North American, European, or East Asian cities, with limited representation of Latin American urban environments. This dataset was assembled to support research on vehicle detection under conditions specific to Brazilian traffic infrastructure — including local vehicle types, road markings, camera hardware/placement, and lighting conditions — and to serve as a benchmark for detection models intended for deployment in this context.