PolarDepth: Polarization-Guided Monocular Depth for Visual Odometry

Naitri Rajyaguru1, Tianfu Wang1, Aryan Tajne1, Botao He1, Jiayi Wu1, Cornelia Fermüller†,1, Christopher Metzler†,1, Yiannis Aloimonos†,1,
1University of Maryland, College Park
IEEE Robotics and Automation Letters (RA-L), 2026
🏆 Best Poster, Insect-Scale Autonomy Workshop, IROS 2026 (attendee-voted) 🥈 Best Poster Runner-Up, Bridging Perspectives in Navigation Workshop, IROS 2026
Paper Video Dataset (releasing soon)

PolarDepth: Left to right: RGB image, RGB-only monocular depth estimate, PolarDepth (ours), and our Polar-RGB polarization encoding. Our method fuses RGB with Polar-RGB, yielding more reliable depth on glass surfaces and improving visual odometry in glass-lined corridors. For visualization, Polar-RGB is shown semi-transparent to reveal the scene beneath.

Abstract

Glass surfaces remain challenging for indoor robot perception. Depth sensors and RGB-only monocular depth estimation often fail because of reflections, refractions, and low-texture regions. To this end, we present PolarDepth, a polarization-enhanced monocular depth framework for glass-dominant environments. We utilize a single polarization sensor to obtain a standard RGB image and a three-channel encoding of polarization cues, designed to be compatible with RGB-trained foundation depth models, and which we call Polar-RGB. This representation enables recovery of structure in transparent and specular regions where RGB cues are unreliable. We predict depth from both the RGB and Polar-RGB representations and fuse the predictions using a learned per-pixel reliability gate for mid-level fusion, highlighting RGB in diffuse regions and polarization on reflective surfaces. We demonstrate improvements in depth estimation and visual odometry performance over RGB-only baselines in glass-walled corridors on both real-world and synthetic data.

-->

Unlocking Depth in Glass

Monocular VO in Glass

Using a polarization information for monocular depth estimation, we enhance odometry accuracy achieving state-of-the-art results on unseen datasets

BibTeX

@ARTICLE{rajyaguru2026polardepth,
  author  = {Rajyaguru, Naitri and Wang, Tianfu and Tajne, Aryan and He, Botao and Wu, Jiayi and Ferm{\"u}ller, Cornelia and Metzler, Christopher and Aloimonos, Yiannis},
  journal = {IEEE Robotics and Automation Letters},
  title   = {PolarDepth: Polarization-Guided Monocular Depth for Visual Odometry},
  year    = {2026},
  volume  = {11},
  number  = {6},
  pages   = {7572--7579},
  doi     = {10.1109/LRA.2026.3685941}
}