M.S. Final Oral Exam: Jiale Feng

Event
Speaker: 
Jiale Feng
Wednesday, January 26, 2022 - 11:00am
Event Type: 

High-fidelity Circuit Boards and Electrical Components Tracking with Traditional and Deep Learning Methods

The thesis examines traditional computer vision and deep learning approaches and presents four architectures for tracking circuit boards and electrical components. The general research addresses assembly tasks on a factory floor, particularly instructing workers. Nowadays, instructions are either printed or available on a monitor. A projector-based system that projects instructions onto the target objects or the workbench is investigated. For that purpose, the target object's category, position, and pose need to be known with high precision.

With that goal in mind, approaches for detection and pose estimation for circuit boards and electrical components are implemented and compared, including feature descriptor-based matching, artificial neural networks (ANNs), and convolutional neural networks (CNNs). The results indicate that ANNs yield the highest precision given the limited training costs. CNNs increase the detection accuracy; however, they require a more extensive training dataset and more time for training.

Each of the architectures proposed in this thesis is designed based on its predecessor, focusing on solving new challenges in the tracking process. Various input data are used and tested, including grayscale images, RGB images, depth maps, and point clouds.

Committee: Soumik Sarkar (co-major professor), Jin Tian (co-major professor), and Adarsh Krishnamurthy

Join on WebEx: https://iastate.webex.com/iastate/j.php?MTID=m7cf67b2b021af00a8f7bc0e1d9...