Seeing Robots definition
Seeing Robots. A central research theme at CVAP is the development of a computer based seeing agent capable of using vision in its interaction with the environment, for e.g. manoeuvring, navigating, grasping and recognizing things. This work is being implemented as "seeing robots" and addresses basic issues on gaze control, attention, figure-ground segmentation, cue integration, and recognition, as well as systems and control issues. A second theme that underpins much of this and other work concerns early vision processing and the computation of scene characteristics. Considerable efforts are being devoted to feature and structure extraction at multiple scales. One of the goals is to develop the notion of a visual-front-end, which can provide a first processing layer for the mentioned seeing agent. perspective and projective transformations can be used to establish image correspondence and to tracking. They also provide a means for model indexing in visual recognition. The latter problem is also studied using other approaches, e.g. based on appearance and also involving learning. The overall goal is to include also these techniques and capabilities in our long-term work on the seeing agent. In addition to these efforts the group is also developing frameworks and computational environments for performing geometric reasoning and analysis as well as for representing geometric objects such as surfaces and volumes. This work has potential applications in computer vision but is also aimed at the study and teaching of geometry as such.