Active Vision in Optical Metrology (Keynote Lecture)
by
Wolfgang Osten
Bremer Institut für Angewandte Strahltechnik, KlagenfurterStr. 2, D-28359 Bremen

For some years a new way of thinking appears with respect to the solution of vision tasks. A paradigm has already arisen, known under the various names Active/Purposive/Animate, etc. ..., Vision. There is an important difference between the active and the classic passive as well as adaptive approach. When we work in passive mode, a set of images is given which have to be processed with the algorithms we are going to develop. On the other hand, when we work in active mode we do not want pre-recorded data since we include the image acquisition as an component of equal importance into the complete evaluation process. And this means far more than making the system adaptive to unfavourable boundary conditions such as a shading correction at unevenly illuminated scene.

Active vision offers a new approach for building intelligent and more flexible systems. In this sense an active system is not only able as just to ßee". It is designed to do something, i.e. to make an action, which is anything that changes the state of the system or the environment. Similar to human perception as an active way to explore a natural scene by changing the view or the focus, the role of the observer and observation system, respectively, is defined as an active component for gathering data. Further on the strategy for solving an image analysis problem is actively influenced by use of controlled feedback loops between the components that capture the optical information and those that evaluate it. In computer vision and optical metrology sometimes the difference between a so called active and passive technique is only made with respect to the way of working with the light, i.e. illumination. For instance in shape measurement the technique shape from shading is classified as an passive technique whereas structured light illumination is called as active. The difference between both approaches is that in the first case the shape is recovered from the brightness at each point of an image usually illuminated by one point light source with the same lighting throughout the surface and in the second case the structure of the lighting is modulated systematically throughout the surface of the object. In my understanding the other way of object illumination by using structured light delivers not more as the possibility to realize the triangulation principle in a field wise manner by addressing each point with a mod 2p phase value. But a lot of problems still remain, especially with respect to the ambiguity of the phase reconstruction. Here the application of more sophisticated procedures such as absolute phase measurement or wavelength scanning is necessary with respect to a unique solution. These methods bring a certain amount of activity into the system since the role of the observer is active by capturing more information as a single view can bring. However, active vision/metrology is not just the use of multiple frames. The modern understanding of being active means that the image acquisition process is controlled and thus constraints that facilitate the recovery of information about the state of the object under test (3D shape, 3D displacement, subsurface defects, etc) are introduced.

The contribution describes the origin of the problems and why only an active approach can be successful in many measurement problems. To explain the different ways of dealing with problems in optical metrology a well known example is used that is quite common for a lot of applications namely the reconstruction of continuous phase distributions from 2D fringe patterns.

Date received: January 28, 1999