Vision Systems to be R.A.R.E or not to be

One could wonder why so few vision systems, while electronic images unfurl and vison is the major human perception capability? The lack of systematic knowledge about image processing algorithmics might provide an explanation. An other might be some lack of constraints to design. Considering R.A.R.E systems (Robust, Autonomous, Rapid, Embarkable) forces first to assemble knowledge and to formalize it. Then transfering it into electronic devices, requires to typify operators: possible lables are primitives like edges, regions, motion ... making a language. Last, the pair (knowledge language) helps designing systems: - Robust: hard to leur, breakdown and aggression resistant, Autonomous; adaptating themselves to new kinds of situations or configurations, - BaDld: extracting pertinent informations to act, at a compatible speed with respect to actions, - Embarkable: bulk, energy consumption and packaging being adapted to the mission; Based on detailed examples in image segmentation and interpretation we illustrate a dedicated programming methodology and environment in vision, together with the way to make systems R.A.R.E.