Austrian Highway Roadwork Views (AHRV) Dataset

Austrian Highway Roadwork Views (AHRV) is a publicly available annotated dataset for object detection in highway roadwork environments. The dataset contains 1,812 scene images with 17,436 bounding-box annotations across 7 road entity classes. The dataset captures real-world variability of Austrian highway construction sites through multiple acquisition campaigns, camera systems, and recording conditions. The collected scenes cover diverse highway environments, lane configurations, viewpoints, lighting conditions, and weather conditions, providing a realistic benchmark for robust roadwork perception. Annotated Object Classes (Bounding Boxes): Traffic signs Chevron signs Traffic cones Traffic beacons Road delineators Construction objects Vehicles The release contains two archives: ahrv_dataset_v100.zip: Dataset files including JPEG images, COCO-format annotations, and dataset documentation. ahrv_benchmark_v100.zip: Benchmark resources including pretrained RF-DETR Large model weights, training and inference scripts, and instructions for reproducing the baseline evaluation. The dataset provides official training, validation, and test splits. Annotations are provided in Microsoft COCO-format to enable compatibility with common computer vision frameworks. The provided RF-DETR Large baseline achieves an instance-weighted mAP₅₀ of 93.5% on the complete test split. AHRV is intended to support research and development in computer vision for highway construction site perception, including applications in advanced driver assistance systems, automated inspection, monitoring, and maintenance. The dataset is released under the CC BY 4.0 license.