Knowledge-based Control of Vision Systems: Design Tools and Case Studies Olena Borzenko a Thesis Submitted to the Faculty of Graduate Studies in Partial Fulfilment of the Requirements for the Degree of Master of Science Graduate Program in Computer Science

To maintain a desired level of system performance in a dynamic environment, a vision system must be able to adapt to a wide range of operating conditions by tuning the parameters that control its multiple processing stages, choosing which vision modules to connect together, and controlling image acquisition. Knowledgebased controllers that reason over explicitly represented knowledge can be used for this task. One drawback with the use of knowledge-based controllers is that it is expensive to collect and represent domain knowledge and to develop different control algorithms. The solution lies in creating a framework and tools for designing intelligent controllers for tasks in dynamic domains. This work presents the INVICON toolkit, based on the IndiGolog agent programming language, for controller design and implementation. It includes a generic domain template, a library of control components, and supporting tools. INVICON has been evaluated in two case studies. Case Study I involves the development of a high-level controller for a vision-based pose estimation system that operates on multiple images acquired under different lighting conditions. The controller heuristically chooses light settings and selects which images to process at each iteration. Case Study II involves improvements to an existing intelligent controller that adjusts edge detection and model-matching parameters of a vision module that estimates the target’s size and position on a single image. The case studies showed that INVICON reduces the effort needed to build intelligent controllers and improves the flexibility and robustness of knowledge-based vision control systems that operate in challenging domains.