The Fraunhofer IKS Safety Net describes our modular approach to AI-based perception systems. According to the demands of the actual scenario, it flexibly combines different detection methods, including object and body-part detection, depth and motion analysis, and out-of-distribution detection, together with a precisely calibrated multi-camera system designed for optimal coverage of the production environment. This redundant architecture makes it possible to reliably identify edge cases in which conventional person detection models often fail, such as unusual poses, partial occlusions, unknown objects, or complex interaction scenarios.
The Fraunhofer IKS Safety Net therefore provides the foundation for demonstrably validating and verifying safe, flexible AI systems in dynamic production environments. It integrates perception, contextual assessment, uncertainty analysis, and safety logic throughout the entire lifecycle—from development and integration to operation.