"Comparing YOLOv8 to YOLOv5 for Pose Estimation Supporting Automated Ae" by Dawson Friesenhahn

Date of Award

3-2024

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Electrical and Computer Engineering

First Advisor

Scott L. Nykl, PhD

Abstract

This work used the newly released YOLO version 8 Object Detection as a feature detector for a monocular pose estimation pipeline and showed up to an 83.6% reduction in translation error when compared to YOLOv5. YOLOv8’s new Pose task’s utility as a feature detector was also explored. It performed consistently worse than the Object Detection task, with typical translation error magnitudes between 3 and 8 times worse than Object Detection. Finally, we investigated the use of constant-sized bounding box labels for Object Detection. Previous approaches have used Bounding Box Corrections (BBC). We found that using constant sized labels increases the number of accurate predictions made by a pose prediction pipeline by up to 11%, while also reducing label generation complexity.

AFIT Designator

AFIT-ENG-MS-24-M-010

Comments

A 12-month embargo was observed for posting this work on AFIT Scholar.

Distribution Statement A, Approved for Public Release. PA case number on file.

Share

COinS