Memorizing Visual Knowledge for Assembly Process Monitoring

Machine learning is a desirable property of computer vision systems. Especially in process monitoring knowledge of temporal context speeds up recognition. Moreover, memorizing earlier results allows to establish qualitative relations between the stages of a processes. In this contribution we present an architecture that learns diierent visual aspects of assemblies. It is organized hierarchically and stores prototypi-cal data from diierent levels of image processing and object recognition. An example underlines that this memory facilitates assembly recognition and recognizes structural relations among complex objects.