Thesis

Advancing utility pole and sign detection through deep learning

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2026
Thesis identifier
  • T18047
Person Identifier (Local)
  • 201675129
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • This thesis addresses the need for safer, faster, and more consistent inspection of wooden utility poles in electrical distribution networks. Manual inspections are costly, labour-intensive, and potentially hazardous, while automated inspection from street-level imagery must cope with background clutter, occlusion, variable lighting, small warning signs, and visually ambiguous wooden poles that may resemble trees or other vertical structures. To address these challenges, this thesis develops and evaluates a deep learning pipeline for automated utility pole detection, warning-sign detection, pole segmentation, and image-plane lean-angle estimation using Google Street View (GSV) imagery. The thesis makes three main contributions. First, it curates the OHL-UK dataset, comprising 4,570 annotated GSV images with bounding boxes, segmentation masks, and lean-angle annotations for wooden utility poles and attached electrical warning signs. Second, it benchmarks established object-detection models, including RetinaNet, You Only Look Once version 3 Tiny (YOLOv3-Tiny), Faster Region-based Convolutional Neural Network (Faster R-CNN), and Detection Transformer (DETR). The DETR-based approach achieved mean average precision (mAP) values of 90.43% for pole detection and 88.26% for warning-sign detection. Third, it extends the DETR-based framework with segmentation capability, enabling pole masks to be used for image-plane lean-angle estimation. On the test set, lean angle was estimated for 1,367 out of 1,433 true-positive poles, with a mean absolute error of 1.01◦. Together, these contributions demonstrate a scalable framework for automated visual assessment of wooden utility poles and provide a reusable dataset and benchmark for future research in electrical infrastructure monitoring.
Advisor / supervisor
  • Di Caterina, Gaetano
Resource Type
DOI

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of thesis T18047 2026-06-17 Public Download