This is an extremely old (1988!) high-level paper by Ruzena Bajcsy.
When I think of "Active Perception", I think of when a robot has sensors and uses information from those to gain a better understanding of its environment. The "active" part implies that the robot must use the sensors in some active way, by performing exploration. (The use of sensors has gotten much more common today, as exemplified by Thomas L. Friedman's book Thank You For Being Late.) Indeed, in the first sentence of the introduction, Bajcsy emphasizes how perception should not be passive. The sensors themselves can collect information passively, but they should be used in an active way. This is basically how our eyes work; they don't modify our environment, but we use then actively to look at important areas. Bajcsy views active perception as "intelligent control theory" involving an agent that reasons and uses prior knowledge.
This paper is a high-level paper, emphasizing the following points:
-
Local Models vs. Global Models. TODO describe ...
-
Example Task 1: 2-D Image Segmentation. TODO describe ...
-
Example Task 2: 3-D Shape Parameterization. TODO describe ...