Available at: https://digitalcommons.calpoly.edu/theses/3412
Date of Award
8-2026
Degree Name
MS in Computer Science
Department/Program
Computer Science
College
College of Engineering
Advisor
Jonathan Ventura
Advisor Department
Computer Science
Advisor College
College of Engineering
Abstract
Three-dimensional cameras provide direct geometric measurements, but their cost, weight, power requirements, and calibration constraints can limit their use in various lightweight or large-scale sensing systems. A potential alternative is to use conventional two-dimensional RGB cameras together with geometric reconstruction models that infer a partial three-dimensional representation from images. This thesis evaluates that possibility for next-best-view (NBV) selection through Sentinel, an occlusion-centered system for static, object-centric scenes with known camera poses and intrinsics. Sentinel converts source RGB observations into pseudo-geometry using monocular depth or point-map predictions, combines those predictions with camera-ray evidence, identifies occluded unknown regions, and selects a candidate view expected to reveal the most previously hidden content. The resulting pseudo-3D state is used for view planning rather than treated as a complete reconstruction, allowing the study to focus on whether inexpensive, widely deployable RGB cameras can provide useful geometric guidance to systems traditionally built around dedicated 3D sensors.
The primary contribution is an evaluation of source-only view-selection quality under a controlled source-only evaluation protocol. Sentinel is compared with random, object-centered, pose-only, and geometry-based baselines on a reproducible synthetic benchmark built from BOP HOPEv2 object models. This evaluation is especially timely because monocular depth and geometric reconstruction have become a rapidly advancing area of computer vision. Sentinel therefore provides a common downstream evaluation for evolving RGB-to-geometry models and examines whether their predictions are sufficiently reliable for next-view decisions.