The development of an active-vision system for 3D-object recognition in robotic assembly workcells was targeted in the CIMLab at the University of Toronto. This system is based on a new approach that combines object pre-marking and active sensing. The objective of this thesis is to address a set of problems within the framework of the proposed active-vision system, namely, optimal object pre-marking, optimal camera-placement, as well as 3D-location estimation of an identified viewed object.

The optimal object pre-marking problem is defined as the determination of the minimum number of markers, and their best locations on a given set of objects. A dynamic-programming-based method is developed to solve the problem. The problem of optimal placement of a mobile camera is addressed in terms of guaranteeing the detection of markers on a viewed object. This problem is divided here into two sub-problems:​ (1) determination of the camera's initial location(s) for detecting an initial marker;​ and (2) determination of the camera's secondary locations for detecting an additional marker, when the first acquired image is insufficient for recognizing the object. The former is solved based on probability theory and computational geometry. The latter is solved by using a method similar to that of the former, as well as utilizing apriori knowledge of the relative locations of markers on the objects under consideration.

The problems of object pre-marking and camera-placement are also integrated into a single problem, where the optimal arrangement of the markers is determined simultaneously with the optimal initial location(s) of the camera.

The commercial CAD package, I-DEAS, is utilized for obtaining the 2D views and the solid-models of 3D objects, to facilitate the solution methods. For the experimental verification of the methods developed, a prototype of the proposed active-vision system is implemented.