Non-Line-of-Sight Imaging via Multi-Bounce Light Transport

This direction is merged into another direction: non-line-of-sight reconstruction. Its last version stays readable and citable here.

Version 1 of 127 papers34 conceptsfrozen 2026-10-04

Imagine trying to see an object hidden behind a wall. To a human eye, the wall is an opaque barrier. But to a computational imaging system, that wall is a potential mirror. While the wall doesn't reflect light perfectly like a silvered glass mirror, it reflects light diffusely, scattering photons in every direction.

The core insight of Non-Line-of-Sight (NLOS) imaging is that these scattered photons carry a "temporal fingerprint" of the hidden scene. By precisely timing the arrival of light that has bounced off a wall, hit a hidden object, and returned to a sensor, researchers can reconstruct the 3D shape of things they cannot directly see.

The Wall as a Virtual Camera

In traditional photography, we rely on a direct line of sight. In NLOS imaging, we treat a visible surface—like a wall or a floor—as a virtual aperture.

As Ahmed Kirmani, Ramesh Raskar, and their colleagues first proposed in Looking around the corner using transient imaging, a diffuse surface can be utilized as a relay. By scanning a laser across different points on a wall, the system effectively creates a synthetic array of light sources and sensors.

Andreas Velten and Ashok Veeraraghavan expanded this in their seminal 2012 work, Recovering three-dimensional shape around a corner using ultrafast time-of-flight imaging. They demonstrated that by treating the wall as a "mirror-less camera," one could solve an "elliptical tomography" problem. Each photon's travel time defines an ellipsoid of possible locations in 3D space; by intersecting thousands of these ellipsoids, the hidden object’s geometry emerges.

Timing the Light Echo

The technical challenge is that light is incredibly fast. To distinguish between a photon that traveled 1 meter and one that traveled 1.01 meters, the system needs picosecond-level accuracy.

Researchers use "transient imaging" to capture these echoes. Systems often employ:

  1. Femtosecond lasers to send out ultra-short pulses of light.
  2. Single-photon detectors or streak cameras to catch the returning "tertiary" reflections (the third bounce).

While early work required vertical walls, Geneviève Gariépy and Jonathan Leach showed in Tracking hidden objects with a single-photon camera that even a horizontal floor can serve as the primary scattering surface. This allows for real-time tracking of moving objects hidden around corners, a capability critical for autonomous vehicle safety.

From Photons to Radio Waves

The principle of multi-bounce reconstruction isn't limited to visible light. Because any wave-like signal follows similar laws of reflection, researchers are applying these concepts to Radio Frequency (RF) and Radar.

In Non-Line-of-Sight 3D Reconstruction with Radar, Haowen Lai and Mingmin Zhao introduced HoloRadar. This system treats environmental surfaces as mirrors for mmWave signals. By "un-mirroring" the reflected data points, they can map hidden structures and people through walls or around corners, which is particularly useful in environments where visual sensors fail due to smoke or lighting. Similarly, Cheng Chen and Takayuki Nishio have explored using Wi-Fi signals (Channel State Information) to "see" people in cluttered industrial spaces.

Neural Inverse Rendering

The most recent shift in the field involves moving away from purely geometric "ellipsoid" math toward neural reconstruction.

Traditional methods often struggle with complex scenes where light bounces more than three times. Modern frameworks like PlatoNeRF, developed by Tzofi Klinghoffer and colleagues, use Neural Radiance Fields (NeRFs) supervised by multi-bounce lidar signals. By modeling the physics of light transport directly within a neural network, these systems can recover complete scene geometry from a single viewpoint without relying on "hallucinations."

As Anagh Malik and Matthew O'Toole demonstrate in Neural Inverse Rendering from Propagating Light, incorporating a time-resolved radiance cache allows the system to efficiently handle both direct and indirect light transport. This enables high-fidelity recovery of not just the shape of hidden objects, but also their material properties.

By turning every surface into a potential sensor, this research direction is effectively removing the concept of "occlusion" from computational vision, allowing machines to perceive the world as a continuous, interconnected web of reflections.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_non_line_of_sight_imaging_via_multi_boun_bf2ba6ad,
  title = {Non-Line-of-Sight Imaging via Multi-Bounce Light Transport},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/non-line-of-sight-imaging-via-multi-bounce-light-transport/rdr_bf2ba6add73d4ae2b8bf20d5c9d99065},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_non_line_of_sight_imaging_via_multi_boun_bf2ba6ad_v1,
  title = {Non-Line-of-Sight Imaging via Multi-Bounce Light Transport},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/non-line-of-sight-imaging-via-multi-bounce-light-transport/rdr_bf2ba6add73d4ae2b8bf20d5c9d99065/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}