Back to search

Monterey, CA; Naval Postgraduate School

POSE ESTIMATION OF UNMANNED AERIAL VEHICLES USING CONVOLUTIONAL NEURAL NETWORKS

Abstract

dc:description.abstract

Unmanned aerial vehicles (UAVs) overwhelm traditional air defense on a cost-per-shot basis. High-energy lasers (HELs) offer a low-cost counter-UAV option, but engagement requires the beam to dwell on an aimpoint, which depends on the orientation of the target. Accurate 3D pose estimation is a prerequisite for the engagement decision. This thesis develops a single-image pipeline that combines an HRNet-W32 convolutional neural network for keypoint detection with a Perspective-n-Point (PnP) algorithm for six degrees of freedom pose recovery, trained on 100,000 synthetic infrared (IR) images of the Mongoose UAV at 4,000 m range. Building on prior NPS work, three contributions follow. First, the left/right wing confusion identified as the primary open problem in that work is resolved with a swap-aware model giving a 4.5x reduction in mean keypoint error over the 10,000 image baseline. Second, the pipeline achieves a median keypoint error of 0.30 pixels, a median rotation error of 0.69 degrees, and a median range error of 11.6 meters, under one percent of the target range. Third, a ground-truth sanity check identifies a geometric ceiling at 78.6 percent on full-pose accuracy. The residual 21.4 percent of viewing angles are ambiguous under near-orthographic monocular projection, and the 100,000 image model saturates this ceiling. The bottleneck is the geometry of the target, not the model. Closing that gap will require multi-frame or multi-view in information, not a better model.

Degree

thesis:*
Name thesis:degree_name
Master of Science in Operations Research
Level thesis:degree_level
Masters
Discipline thesis:degree_discipline
Operations Research
Department dc:contributor.department
Operations Research (OR)
Grantor dc:publisher
Monterey, CA; Naval Postgraduate School
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Atiles, Ricardo H.
Advisors dc:contributor.advisor
  • Agrawal, Brij N.
  • Smith, Matthew D.
Committee member dc:contributor.committeemember
  • Herrera, Leonardo E.

Rights

dc:rights
Statement dc:rights
  • This publication is a work of the U.S. Government as defined in Title 17, United States Code, Section 101. Copyright protection is not available for this work in the United States.
  • Distribution Statement A. Approved for public release: Distribution is unlimited.

Identifiers

dc:identifier.*
Dc Identifier Programcode
361, Operations Research Logistics Analysis
Thesis Thesis Identifier
42174

Chain of custody

source
Harvested from
Naval Postgraduate School
Base URL
calhoun.nps.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Atiles, Ricardo H.. POSE ESTIMATION OF UNMANNED AERIAL VEHICLES USING CONVOLUTIONAL NEURAL NETWORKS. Masters thesis, Monterey, CA; Naval Postgraduate School, 2026. https://hdl.handle.net/10945/75656