Date of Award

12-2015

Document Type

Thesis

Degree Name

Master of Science (MS)

Legacy Department

Electrical Engineering

Advisor

Hoover, Adam W

Committee Member

Groff, Richard E

Committee Member

Wang, Yongqiang

Abstract

This thesis considers the problem of assisting a human user setting up an automated Visual Inspection (VI) system. The VI system uses a stationary camera on an automobile assembly line to inspect cars as they pass by. The inspection process is intended to identify when parts have been missed or incorrect parts have been assembled. The result is reported to a human working on the assembly line who then can take corrective actions. As originally developed, the system requires a setup phase in which the human user places the camera and records a video of at least 30 minutes length to use for training the system. Training includes specifying regions of cars passing by that are to be inspected. After deployment of a number of systems, it was learned that users could benefit from being provided guidance in best practices to delineate training data. It was also learned that users could benefit from simple visual feedback to ascertain whether or not an inspection problem was suitable for a VI system or if the problem was too challenging. This thesis describes a few methods and a new software tool intended to address this need.

Share

COinS