Polarization Image Fusion
for Extreme Lighting
A semester-long evaluation of polarization imaging as an input to visual SLAM — testing whether combining polarization cues with intensity gives ORB-SLAM3 a more robust input than intensity alone, under the saturation and harsh lighting of the lunar surface.
Project Overview
The Moon has no atmosphere, so surfaces are lit by direct sunlight with no scattered fill light. This produces hard shadows, saturated highlights on regolith, and large appearance changes as the Sun moves. Visual SLAM depends on repeatable image features, and these conditions remove them.
The cost of this weakness is real: Ingenuity's January 2024 navigation failure on Mars has been linked to wrongly detected terrain features (Gonzalez et al., 2026).
Polarization depends on surface orientation and material, so it may preserve structure where intensity has clipped. The one prior study at QUT's Yandiwanba Lunar Testbed ran four VSLAM systems on polarization camera sequences and reported their limits under glare and perceptual aliasing. This project asks a narrower question: does fusing polarization cues with intensity, before feature extraction, give ORB-SLAM3 a better input than intensity alone?
The project covers method, results and deliverables, as well as risk, ethics, sustainability and stakeholder input.
Why Lunar Lighting Breaks Visual SLAM
Without an atmosphere, the conditions visual SLAM relies on disappear — and so does the skylight cue most polarization navigation uses.
No Fill Light
- No atmosphere, so no scattered light
- Surfaces lit by direct sunlight only
- Hard shadows hide terrain structure
Saturated Highlights
- Bright regolith clips the sensor
- Intensity texture is lost where pixels saturate
- Polarization may preserve that structure
Moving Sun
- Large appearance changes as the Sun moves
- Visual SLAM needs repeatable features
- The same place can look entirely different
No Skylight Compass
- Most polarization navigation uses skylight as a compass
- That pattern is created by the atmosphere
- On the Moon, only ground-reflected polarization is usable
From Pixels to Polarization
A division-of-focal-plane camera places a 2 × 2 grid of linear polarizers over its pixels. The four analyser images give the Stokes parameters, and from them the degree and angle of linear polarization.
Fusion Pipeline
Fusion happens at the image level, before feature extraction. ORB-SLAM3 itself is left unchanged and compared against an intensity-only (S₀) baseline.
Study Scope
A deliberately controlled comparison — only the input image changes.
Data — limited to the polarization recordings supplied to the project.
SLAM system — monocular ORB-SLAM3.
Fusion — performed at the image level, before feature extraction.
SLAM internals — not modified; the comparison isolates the effect of the input image.
Report Structure
Beyond method and results, the report covers deliverables, risk, ethics, sustainability, standards and stakeholder input.
Data, Pipeline & Protocol
The supplied recordings, the fusion pipeline and the evaluation protocol.
Results
Findings reported in three stages.
Deliverables & Challenges
What the semester produced and the obstacles along the way.
Professional Context
Risk, ethics, sustainability, standards and stakeholders.
Conclusion
Conclusions with recommendations for future work.