Brazilian highways dataset - satellite images

The Brazilian Highways dataset was developed to meet the growing demand for data that represents the reality of Brazilian highways, particularly in vehicle detection in satellite images. This dataset contains images from eight Brazilian highways (SP 160, SP 150, BR 101, BR 116, SP 021, SP 041, SP 070, SP 280) high-quality RGB images. Each image includes detailed annotations in two main classes—light vehicles and heavy vehicles—with a ground sampling distance (GSD) of 0.074 m/pixel and 0.3 m/pixel, providing essential data for training deep learning models in computer vision. This dataset is available and offers an unprecedented and valuable resource for developing technologies adapted to the characteristics of the Brazilian fleet and highways.

The dataset development was conducted under the guidance of Prof. Dr. André Luiz Cunha, with the active participation of researchers Luan André Contel and Crhistian Emilio Ribeiro. The project stands out for its relevance in the field of vehicle detection by satellite images through computer vision and deep learning techniques and its potential applications in urban planning and traffic management.

Funded by the University of São Paulo (USP), through the PUB scholarship program, the work represents a significant advance in the use of technologies for vehicle detection for innovative solutions in urban mobility. The initiative not only contributes to the academic development of those involved but also to technological progress in the area of intelligent transport systems.