SP-LiDAR for Fast and Robust Depth Imaging at Low SBR and Few Photons

Single photon LiDAR has demonstrated remarkable proficiency in long-range sensing under conditions of weak returns. However, in the few-photon regime (SPPP ≈ 1) and at low signal-to-background ratios (SBR ≤ 0.1), depth estimation is subject to significant degradation due to Poisson fluctuations and background contamination. To address these challenges, we propose GLARE-Depth, a patch-wise Poisson-GLRT framework with reflectance-guided spatial fusion. In the temporal domain, our method employs a continuous-time Poisson-GLRT peak search with a physically consistent exponentially modified Gaussian (EMG) kernel, complemented by closed-form amplitude updates and mode-bias correction. In the spatial domain, we implement a methodology that incorporates reflectance-guided, edge-preserving aggregation and confidence-gated lightweight hole filling to enhance effective coverage for few-photon pixels. In controlled simulations derived from the Middlebury dataset, under high-background conditions (SPPP ≈ 1, SBR ≈ 0.06–0.10), GLARE-Depth demonstrates substantial gains over representative baselines in RMSE, MAE, and valid-pixel ratio (insert concrete numbers when finalized) while maintaining smoothness in planar regions and sharpness at geometric boundaries. These results highlight the robustness of GLARE-Depth and its practical potential for low-SBR scenarios.