A Neural Reconstruction Method for Non-Line-of-Sight Imaging

Abstract

Non-Line-of-Sight Imaging (NLOS) addresses the challenge of capturing a scene which is not directly visible from a sensor line-of-sight, obscured by an occluder. This work uses time-of-flight formulation to propose a learning-based method that represents implicitly the hidden scene partially disambiguated by transient imaging and explores new visibility features coherent with this representation.