Dataset for Hand Detection Near a Sliding Table Saw in Safety-Critical Woodworking Scenarios

This dataset contains image and annotation data for detecting hands in the working area of a sliding table saw (Formatkreissäge). Its purpose is to support the development and evaluation of AI models that identify potentially dangerous hand positions near the saw blade and can serve as a basis for automated machine-stop systems. The data were recorded at Fraunhofer IPA in a realistic laboratory environment using an overhead camera mounted above the saw. The recordings cover several defined safety-critical sawing scenarios, including concealed cuts, corrective cuts, rip cuts, groove cuts, and operations involving a push stick. The dataset also includes variations with and without gloves in order to capture different visual appearances of hands and improve the robustness of detection models. The manually verified subset contains 12,183 individual images with bounding-box annotations in YOLO format for the class Hand. Of these, 5,782 images show hands wearing gloves and 6,401 images show hands without gloves. The dataset is split into training, validation, and test sets (6,507 / 2,085 / 3,591 images). The split is person-based in order to avoid data leakage between subsets: training uses Persons 2, 4, and 5, validation uses Person 1, and testing uses Person 3. This dataset is particularly suitable for research in object detection, safety-critical AI, human-machine interaction, and occupational safety in woodworking environments.