The subproject "Perception of Human" (B04) deals with video-based recognition in the context of manual disassembly/reassembly/assembly in the circular factory. The goal is to develop deep-learning-based methods that capture the current state of humans at a fine granular level, distinguish between known and unknown situations, and generalize to new sequences of actions with little training data. To train the models, an annotated database will be created in close collaboration with other project partners. Recognition pipelines for additional features (e.g., body pose) that are needed by other subprojects will also be developed.

Principal Investigators