TRANSSET: A drone-captured dataset for enhancing vehicle detection and traffic monitoring algorithms
TL;DR
This dataset addresses the scarcity of high-quality aerial traffic data by offering diverse viewing angles, exposures, and backgrounds and serves as a vital resource for training, validating, and testing machine learning models, particularly object detection algorithms.
Abstract
This article presents TRANSSET, a comprehensive dataset of drone-captured imagery designed to advance traffic detection and vehicle classification research. The dataset consists of over 4704 high-resolution images extracted from 4 K video footage collected at two highway locations in North Carolina, USA (Interstate 40 and Interstate 440). Data acquisition was performed using a DJI T600 multi-rotor drone hovering at altitudes between 140 and 200 feet, providing unique oblique perspectives of traffic flow under varying weather conditions. Each image is meticulously annotated with bounding boxes and class labels utilizing a comprehensive eight-class vehicle taxonomy (sedan, SUV, pickup truck, van, car, truck, trailer, hatchback) in three widely used formats: PASCAL VOC XML, COCO JSON, and TXT. This dataset addresses the scarcity of high-quality aerial traffic data by offering diverse viewing angles, exposures, and backgrounds. It serves as a vital resource for training, validating, and testing machine learning models, particularly object detection algorithms, thereby supporting the development of robust intelligent transportation systems and automated traffic monitoring solutions.