Abstract

Transient non-line-of-sight imaging techniques reconstruct hidden scenes by analyzing the time of flight of light scattered off a visible secondary surface, or relay wall. Despite many promising approaches, all face the inherent problem of the missing cone, which restricts surface visibility based on their position and orientations relative to the relay wall. Drawing inspiration from stereo technologies from computer vision, we devise a setup consisting of two distinct relay walls. We leverage Phasor Fields that computationally model both relay walls as generalized virtual camera apertures. This approach allows us to combine the contributions from each relay wall, including the signal obtained by illuminating one wall and capturing the other, information that would be lost otherwise. Our results demonstrate that our proposal diminishes the effect of the missing cone by making the problem better posed. Additionally, by analyzing the visibility conditions of the missing cone, we extract orientation cues from each relay wall contribution. We use this information to enhance visualizations.

Results

Our setup allows for analyzing the contribution of the different relay walls to the scene, that we color code in the hidden scene reconstructions. Next, we show the contribution variation depending on the angle between relay walls (only left relay wall rotates).

Results Results

Paper

Paper: PDF

Bibtex

@article{luesia2026stereonlos, author={Luesia-Lahoz, Pablo and Cartiel, Sergio and Mu{\~{n}}oz, Adolfo}, title={Stereo non-line-of-sight imaging}, journal={The Visual Computer}, year={2026}, month={Jan}, day={29}, volume={42}, number={2}, pages={148}, abstract={Transient non-line-of-sight imaging techniques reconstruct hidden scenes by analyzing the time of flight of light scattered off a visible secondary surface, or relay wall. Despite many promising approaches, all face the inherent problem of the missing cone, which restricts surface visibility based on their position and orientations relative to the relay wall. Drawing inspiration from stereo technologies from computer vision, we devise a setup consisting of two distinct relay walls. We leverage phasor fields that computationally model both relay walls as generalized virtual camera apertures. This approach allows us to combine the contributions from each relay wall, including the signal obtained by illuminating one wall and capturing the other, information that would be lost otherwise. Our results demonstrate that our proposal diminishes the effect of the missing cone by making the problem better posed. Additionally, by analyzing the visibility conditions of the missing cone, we extract orientation cues from each relay wall contribution. We use this information to enhance visualizations.}, issn={1432-2315}, doi={10.1007/s00371-025-04340-7}, url={https://doi.org/10.1007/s00371-025-04340-7} }

Acknowledgments

We thank the people from the Graphics and Imaging Lab for their support, and especially Jorge García, Santiago Jiménez and Julio Marco for their assistance. This work was funded by HORIZON EUROPE Research and Innovation Actions (101070310), and Ministerio de Ciencia, Innovación y Universidades/Agencia Estatal de Investigación/10.13039/501100011033 (Project PID2019-105004GB-I00).