This thesis presents a machine vision system for supervision and guidance of a tube inspection manipulator used for eddy-current inspections of steam generators in nuclear power plants. In one inspection cycle, few hundred or even few thousand steam generator tubes need to be tested, so omissions and errors can occur. A supervisory vision system is introduced into the inspection manipulator motion control loop to increase its reliability and eliminate the problem of tube misidentification. The system is using a smart camera, which performs both image acquisition and analysis, mounted at the manipulator instrument platform and linked via Ethernet connection to the motion controller and to the remote operator station. The acquired image is processed to locate and track circular tube openings. By tracking manipulator movements over the tube mesh the system can identify current tube coordinates and verify that the eddy-current probe is positioned into the desired tube. On final approach of the manipulator arm to the tube opening, the vision system can automatically guide the manipulator to achieve accurate positioning for probe insertion, thus eliminating the need for operator input and reducing probe wear. Integration of the developed vision system into the manipulator motion control system enables development of a more reliable and a fully automated inspection system to meet the ever-increasing nuclear power plant safety standards. The presented system was developed and tested on a laboratory model and then successfully implemented on a real manipulator during an actual inspection in a nuclear power plant.