Computational single-photon depth imaging without transverse regularization

Abstract

Depth profile reconstruction of a scene at low light levels using an active imaging setup has wide-ranging applications in remote sensing. In such low-light imaging scenarios, single-photon detectors are employed to time-resolve individual photon detections. However, even with single-photon detectors, current frameworks are limited to using hundreds of photon detections at each pixel to mitigate Poisson noise inherent in light detection. In this paper, we discuss two pixelwise imaging frameworks that allow accurate reconstruction of depth profiles using small numbers of photon detections. The first framework addresses the problem of depth reconstruction of an opaque target, in which it is assumed that each pixel contains exactly one reflector. The second framework addresses the problem of reconstructing multiple-depth pixels. In each scenario, our framework achieves photon efficiency by combining accurate statistics for individual photon detections with a longitudinal sparsity constraint tailored to the imaging problem. We demonstrate the photon efficiencies of our frameworks by comparing them with conventional imagers that use more naïve models based on high light-level assumptions.

Signal Processing Based Pile-up Compensation for Gated Single-Photon Avalanche Diodes

“…Kirmani et al [28] proposed a low-photon-flux imaging system that can reconstruct both depth and reflectivity of a scene from the first arriving photons. Shin et al [29,30] extended this system that can reproduce similar results even with a nano-second jitter system but with prior knowledge on the scene. Gariepy et al [4] used a 32×32 SPAD array to capture transient images directly at 67 ps temporal resolution.…”

Restoration of multilayered single-photon 3D Lidar images

“…applied on gated histograms Y Man . It is also compared to the recently proposed approach [4,22], which assumes an 1 regularization term, given by…”

Robust Restoration of Sparse Multidimensional Single-Photon LiDAR Images

“…This section presents the regularization term associated with the data support. As highlighted in [5], [27], 3D Lidar data is sparse especially in the low acquisition time regime. However, considering sparsity alone does not help separate the target's returns from those due to the background noise.…”

“…A possible strategy to deal with this is to impose this depth localization locally, i.e., for each small patch of pixels. In this paper, we choose to combine the priors in [5], [27] Illustrative examples of the effect of different support regularizations, i.e., (left) effect of the 1 -based regularization in [5], [27], (middle) the 2,1based regularization in [25], and (right) the 2,1 -based regularization proposed in this paper. The black cubes represent the obtained histograms, the dots represent the detected photons (red for the target and yellow for the background returns) and the green cubes represent the detected supports promoted by the regularizations.…”