Towards Automatic VR Scene Generation for Safety Training Scenarios Using Manually Annotated 3D Object and Spatial Relation Data

Key Points

• The aim is to generate virtual reality safety training scenarios using annotated data from accident reports.
• Collection of 100 manually annotated accident reports.
• Integration of deep learning models with a safety library for scene generation.
• The dataset enables the development of realistic VR training scenarios.
• Shows potential for enhancing safety training efficiency through automation.