Date of Award
1-2024
Document Type
Thesis
Degree Name
Master of Science (MS)
College/School
College of Science and Mathematics
Department/Program
School of Computing
Thesis Sponsor/Dissertation Chair/Project Chair
Weitian Wang
Committee Member
Hao Liu
Committee Member
Rui Li
Abstract
This work presents a comparative analysis of feature-matching techniques implemented on low-end hardware, focusing on their efficiency and performance under various image transformations. The study evaluates several well-established feature matching algorithms, including ORB, AKAZE, BRISK, FAST combined with ORB, and SIFT, for their robustness against rotation, perspective, and scale changes in images. The base image used for experimentation features is the Montclair State University's Red Hawk mascot—a complex, textured subject that presents a substantial challenge for feature matching algorithms. The experiment simulates real-world conditions by applying a series of transformations to the base image and utilizes the default settings of the chosen methods to ensure a fair comparison within the constraints of a Raspberry Pi 4 Model B environment. Performance is measured regarding reprojection error, computation time, and memory usage. The study establishes a baseline by comparing two identical images, providing a reference for the most efficient scenario. The results demonstrate that while ORB consistently exhibits the fastest processing time across all transformations, it sometimes fails to find adequate good matches under complex conditions. AKAZE and BRISK balance speed and precision, proving stable across various transformations. The study contributes to understanding feature matching in constrained computing environments, highlighting the trade-offs between efficiency and accuracy. The findings underscore the potential of low-cost hardware for robust computer vision applications and pave the way for advancements in embedded and real-time vision systems.
File Format
Recommended Citation
Du, Xu, "Feature Matching Methods Comparison with Limited Computing Power" (2024). Theses, Dissertations and Culminating Projects. 1381.
https://digitalcommons.montclair.edu/etd/1381