As road computer vision technologies evolve, the need for a more comprehensive data that captures various infrastructure classes becomes crucial for safe deployment of automated vehicles. This paper proposes RoadVision, which is a comprehensive dataset that provides a well-structured collection of 4,487 high-resolution images annotated for 25 distinct labels and designed for object detection in road environments. The images were sourced from urban and suburban areas to ensure diversity and realism. Annotations were created with a high level of precision, capturing essential details such as bounding box coordinates, object types, and class labels. The dataset includes a wide range of traffic-related elements, such as traffic signs (e.g., speed limits, no stopping signs), road objects (e.g., traffic cones, pedestrian crossings), and road markings, which particularly appears to lack in many other datasets. Dataset was accurately annotated to be suited for training and evaluating AI-based object detection and classification models, with applications in autonomous driving, traffic safety, International Road Assessment Program (iRAP), and intelligent transportation systems (ITS). The dataset is hosted as open access on Roboflow Universe, offering multiple annotation formats (e.g., CSV, JSON, XML, and TXT) to ensure compatibility with different machine learning and AI frameworks. We tested the dataset using YOLOv11 to ensure its applicability. We found that results were promising reaching about 97% precision for some classes. The dataset sets a new standard for object recognition research and is available at: https://universe.roboflow.com/detection-flj91/irap-zcrw6-uif6x