In this introduction we have presented an approach for extraction and tracking of contours. We have formally introduced active rays, a new representation of 2D contours based on the ideas of active contours. An internal energy as well as an external energy are defined and allows for the extraction of a contour by means of an energy minimization step. The reduction of the contour point search from the 2D image plane to 1D signals allows for the definition of texture energies on 1D signals. Thus, a fast and --- as shown in the experimental part --- accurate contour extraction is possible, even for objects, where the gradient is not suited for contour point localization. The quantitative evaluation of moments of different order has shown, that there is not one judgement function, which is able to correctly extract arbitrary contours. Indeed, we have found one judgement function for each image out of the database of 100 images, which is suitable to extract the contour of the circle.
Thus, it is natural to extend the approach to a combination of several judgement functions. For a given problem domain such a combination might be automatically built by having a classified training set and applying genetic algorithm to find the optimal combination. This appears to be also a good approach to automatically compute other parameters of active rays, like the window width, the number of hypotheses, etc.
For real--time application, this computational inexpensive approach
can be an additional knowledge source to other contour based tracking
algorithms. It is conceivable
to combine this method with the approach of [18], who
computes also a radial representation of the object based on
the optical flow. The use on segmentation problems, for example
in the area of medical images, must be evaluated separately. Similar to
active contours, the accuracy of the contour extraction depends
on the chosen parameter 
and 
.