Robot Vision System Using Differential Image Information And Geometrical Relations Of Moving Edges

For robot vision, it is important to extract features from a sequence of image frames rather than from each image individually. The high rate of image frames in a video sequence supplies a high amount of information. Extracting this information makes it necessary to introduce several processing stages which exchange data asynchronously and rhus working at different speed. Algorithms in the early processing stage have to be fast in order to work on each frame whereas algorithms in later stages which extract information on a hight:r level may be slower and more complex. For this paper, it is assumed that the camera can be moved in a controlled way. A structure of a robot vision system is presented which detects the movements of edges in a sequence of images and assigns the third dimension by interpreting the movements. It considers edges due to the static environment and the movement of the camera and edges due to moving: objects. A last stage deals with the object recognition. The whole system is supervised by a "Vision Manager" which controls the camera as well as the regions of interest and thus ensures that important tasks are carried out within a limited range of processing power.