Technology has to a large extent driven the progress and it has gradually enabled use/formulation of more advanced theories for problems that earlier were considered out of reach.

Commitment to science is critical to achieve success. This at the same time requires organisation as progress otherwise might end up being incremental without major breakthroughs. Without organisation there is at the same time a risk that the same methods are re-invented at a later stage. For the evaluation of progress and results it is also essential to be explicit about ones reference. Computer vision includes both biological and engineering goals and it is important to specify towards which field the research is directed at it determines how one should evaluate the results. It was emphasised that it is essential to stick to a goal/topic even if it implies switching between funding agencies.

It is essential that the task / system is considered from the start. This also implies that a multi-disciplinary approach must be used. Computer vision should exploit existing theories and techniques from other sciences. This includes fields like physics, biology, computer science, artificial intelligence, statistics and control engineering. At the same time there is a conflict with the disciplines as each of these disciplines in turn require credit for progress, both due to funding constraints and due to interdisciplinary collaborations. This requires a delicate balance between other disciplines and computer vision.

Time/dynamics has only recently been recognised as an important aspect of building operational systems. This is partly due to the fact that construction of fully operational system only recently has become possible. In addition methods for description of dynamics at several different levels from control theory to temporal logic have only recently been integrated into a coherent framework.

The combination of different disciplines has only happened recently which in part is due to the fact that there has been a kind of 'religious' separation between fields like geometry, pattern recognition, control theory and artificial intelligence. I.e., simple applications, for example in pattern recognition, were not considered computer vision. In the view of complete systems it is, however, now apparent that such systems can only be built when the disciplines are combined with proper use of a multi-disciplinary approach.

Computer vision should have clear goals and they should at the same time be meaningful to science in general. This is in particular important to motivate computer vision as a discipline. Other sciences have defined golden standards and use hypothesis testing etc. as a basis for their work. Such standards should also be imposed on scientific journals and conferences as it will allow broader recognition of computer vision as a well-established science. This at the same time implies that performance characterisation becomes a critical topic for evaluation of developed techniques.

The issue of adequate computer power was discussed. It is not immediately obvious if we have enough computing power to solve current problems. A more important problem might, however, be adequate knowledge. Most systems developed today use little or no explicit knowledge. Another related problem is that almost no systems have a well characterised knowledge base, which implies that the systems can not be combined with methods for learning and/or adaptation.