An increasing demand for high-performance and lower-cost LiDAR has led to a wealth of new research and commercial technologies utilizing photon-counting detection and image reconstruction techniques. In this work we demonstrate results from a compact and portable photon-counting LiDAR prototype, consisting of a high-speed digital-micromirror-device, short-pulsed infrared laser, photon-counting photomultiplier and FPGA-based TCSPC electronics. We evaluate the system performance when operating at ranges of up to 20m, using different scanning and reconstruction techniques which employ compressed sensing to increase the frame rate. Deep neural networks are computational models for learning representations of data with multiple levels of abstraction. Recently, there has been interest in using deep neural networks as a promising alternative to traditional compressive sensing techniques. In this work we will demonstrate progress made in using a deep convolutional auto-encoder network for recovering 3D images from a photon-counting LiDAR, which provides a computationally-efficient pipeline for solving underdetermined problems with better quality, in real-time. We anticipate that low-cost photon-counting LiDAR, enriched by deep-learning, will play an important role in many commercial sensing applications such as autonomous vehicles.

© (2018) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.

Matthew P. Edgar, Miles Padgett, Catherine Higham, and Roderick Murray-Smith, "Real-time photon-counting LiDAR enhanced with deep-learning (Conference Presentation)," Proc. SPIE 10799, Emerging Imaging and Sensing Technologies for Security and Defence III; and Unmanned Sensors, Systems, and Countermeasures, 107990B (Presented at SPIE Security + Defence: September 12, 2018; Published: 11 October 2018); https://doi.org/10.1117/12.2503325.5847436275001.