While object detection has achieved remarkable success in tasks such as autonomous driving and intelligent transportation, there is still a gap in meeting the fine-grained detection requirements for specialized road traffic survey tasks. To address this, we built a dedicated vehicle detection dataset tailored for traffic surveys of roads, named the VDD4TS. Within monitoring scenarios, VDD4TS aims to perform detection research on ten vehicle categories, including sedans, small trucks, buses, medium-sized trucks, large trucks, extra-large vehicles, container trucks, tractors, motorcycles, and electric cars. It comprises 12,535 high-quality images, encompassing 59,242 bounding boxes. It accounts for diverse lighting and weather conditions. Furthermore, the dataset possesses three noteworthy attributes: more detailed vehicle classification, a specific focus on suburban and rural scenes that road network detection systems relatively less represent, and distinct and informative annotation details. Extensive baseline experiments were conducted in object detection tasks to validate the dataset's effectiveness. The experimental results indicate that the VDD4TS dataset poses challenges for current mainstream object detection algorithms and is suitable for evaluating more complex advanced algorithmic models.
VDD4TS: A Vehicle Detection Dataset for Traffic Survey of Road
2023-11-03
4085084 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Road traffic: Vehicle detection and classification
IEEE | 2017
|R2D2: Rural Road Detection Dataset
IEEE | 2023
|Engineering Index Backfile | 1938