Photon-Efficient Non-Line-of-Sight Imaging

Non-line-of-sight (NLOS) imaging techniques have the ability to look around corners, which attracts growing interest for diverse applications in autonomous navigation, medicine, transportation, manufacturing and many other domains. At present, to recover the hidden scenes, most existing transient NLOS methods need full histogram at each scanning point, which requires hund reds of detected photons to obta in both the time-of-flight (TOF) information and the intensity information. In this paper, we introduce a photon-efficient method to recover the hidden scene using only one detected photon, which contains only the TOF information of the detected photon, at each scanning point. Our method first uses the single detected photon to estimate the intensity information, and then introduces a convex optimization method with a tailored joint regularization term to recover the 3D information of the hidden scene. The regularization term contains a non-local self-similarity (NLSS) norm, which is used to capture the local structure of the hidden scene, and a total variation (TV) semi norm, which is used to enhance the edge features. To evaluate the performance of our method, both simulations and experiments are demonstrated in this paper. The results show that this photon-efficient method outperforms previous approaches under low-flux conditions.

Automated summary

Non-line-of-sight (NLOS) imaging techniques have the ability to look around corners, which attracts growing interest for diverse applications. This paper proposes a photon-efficient method to recover hidden scenes using only one detected photon at each scanning point. The method estimates the intensity information and utilizes convex optimization with joint regularization terms to recover the 3D information of the hidden scene. Simulations and experiments show that this method outperforms previous approaches under low-flux conditions.