Parallel Block Compressive LiDAR Imaging

Abstract

We propose an architecture for reconstructing depth images from raw photon count data. The architecture uses very sparse illumination patterns, making it not only computationally efficient, but due to the significant reduction in illumination density, also low power. The main idea is to apply compressive sensing (CS) techniques to block (or patch) regions in the array, which results in improved reconstruction performance, fast concurrent processing, and scalable spatial resolution. Using real and simulated arrayed LiDAR data, our experiments show that the proposed framework achieves excellent depth resolution for a wide range of operating distances and outperforms previous algorithms for depth reconstruction from photon count data in both accuracy and computational complexity. This enables eye-safe reconstruction of high-resolution depth maps at high frame rates, with reduced power and memory requirements. It is possible to sample and reconstruct a depth map in just 12 ms, enabling real-time applications at frame rates above 80 Hz.