With increasing computational power and decreasing size, computers nowadays are already wearable and mobile. They become attendant of peoples' everyday life. Personal digital assistants and mobile phones equipped with adequate software gain a lot of interest in public, although the functionality they provide in terms of assistance is little more than a mobile databases for appointments, addresses, to-do lists and photos. Compared to the assistance a human can provide, such systems are hardly to call real assistants.

The motivation to construct more human-like assistance systems that develop a certain level of cognitive capabilities leads to the exploration of two central paradigms in this work. The first paradigm is termed cognitive vision systems. Such systems take human cognition as a design principle of underlying concepts and develop learning and adaptation capabilities to be more flexible in their application. They are embodied, active, and situated. Second, the ego-vision paradigm is introduced as a very tight interaction scheme between a user and a computer system that especially eases close collaboration and assistance between these two. Ego-vision systems (EVS) take a user's (visual) perspective and integrate the human in the system's processing loop by means of a shared perception and augmented reality. EVSs adopt techniques of cognitive vision to identify objects, interpret actions, and understand the user's visual perception. And they articulate their knowledge and interpretation by means of augmentations of the user's own view.

These two paradigms are studied as rather general concepts, but always with the goal in mind to realize more flexible assistance systems that closely collaborate with its users. This work provides three major contributions. First, a definition and explanation of ego-vision as a novel paradigm is given. Benefits and challenges of this paradigm are discussed as well. Second, a configuration of different approaches that permit an ego-vision system to perceive its environment and its user is presented in terms of object and action recognition, head gesture recognition, and mosaicing. These account for the specific challenges identified for ego-vision systems, whose perception capabilities are based on wearable sensors only. Finally, a visual active memory (VAM) is introduced as a flexible conceptual architecture for cognitive vision systems in general, and for assistance systems in particular. It adopts principles of human cognition to develop a representation for information stored in this memory. So-called memory processes continuously analyze, modify, and extend the content of this VAM. The functionality of the integrated system emerges from their coordinated interplay of these memory processes.

An integrated assistance system applying the approaches and concepts outlined before is implemented on the basis of the visual active memory. The system architecture is discussed and some exemplary processing paths in this system are presented and discussed. It assists users in object manipulation tasks and has reached a maturity level that allows to conduct user studies. Quantitative results of different integrated memory processes are as well presented as an assessment of the interactive system by means of these user studies.