Back to search

University of Southampton

Predicting high-resolution terrain properties in robotic survey applications

Abstract

dc:description.abstract

In this work we present a novel framework for predicting high-resolution terrain observations applicable to aerial, underwater and planetary robotic platforms. We developed a hybrid system architecture capable of predicting high resolution properties of the terrain from widely available lower resolution priors. Our architecture seamlessly integrates state-of-the-art unsupervised feature extraction techniques with a robust predictive engine based on Bayesian machine learning. Spatial autocorrelation of low resolution priors is used as a principled way to optimize the search of the distance-related parameters that maximizes quality of the predictions. By using this approach, we effectively narrowed the search range of distance parameters to between 10.1% and 31.3% of the original exploration range, substantially reducing the computational cost associated with feature extraction optimization, when compared to a naive exploratory approach. Information about the expected distribution of high-resolution derivatives is used to improve the predictive engine accuracy in a generalizable fashion. The impact of different label aggregation strategies is measured, with an improvement of 32.9% in the habitat class proportion when using a representation compatible with Bayesian models, also outperforming the accuracy of published results in the same area of study for seafloor classification task by a margin of 17.8% while using less input data. Our proposed solution is capable of predicting terrain slope for vehicle landability and habitat classes distribution across different types of terrain data and resolutions with minimal human intervention. We demonstrate the effectiveness of our approach in three different environments: (i) underwater (ii) terrestrial and (iii) planetary surfaces.

Degree

thesis:*
Name dc:type.qualificationname
Ph.D.
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Southampton
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Cappelletto, Jose De La Cruz
Advisors dc:contributor.advisor
  • Thornton, Blair
  • White, David
  • Newborough, Darryl
  • Dix, Justin
ORCID person:identifier.orcid

Rights

Language dc:language
English, eng

Chain of custody

source
Harvested from
University of Southampton
Base URL
eprints.soton.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Cappelletto, Jose De La Cruz. Predicting high-resolution terrain properties in robotic survey applications. doctoral thesis, University of Southampton, 2025.