Fast Non-Line-Of-Sight Imaging with Two-Step Deep Remapping
Dayu Zhu, Wenshan Cai
Abstract
Conventional imaging only records the photons directly sent from the object to the detector, whereas non-line-of-sight (NLOS) imaging takes the indirect light into account. Most NLOS solutions employ a transient scanning process, followed by a physical-based algorithm to reconstruct the NLOS scenes. However, transient detection requires sophisticated apparatus, long scanning time, and low robustness to the ambient environment, and the reconstruction algorithms are typically time consuming and computationally expensive. Here, we propose a new NLOS solution with innovations on both equipment and algorithm. We apply inexpensive Lidar for detection, with much higher scanning speed and better compatibility to real-world imaging. Our reconstruction framework is deep learning based, with generative two-step remapping strategy to guarantee high reconstruction fidelity. The overall detection and reconstruction process allows for millisecond responses, with state-of-the-art reconstruction performance. We have experimentally tested the proposed solution on both synthetic and real objects and further demonstrated our method to be applicable for full-color NLOS imaging.