Structure-guided adaptive total variation for parameter-free passive non-line-of-sight imaging

Non-line-of-sight (NLOS) imaging aims to recover hidden scenes from indirect reflections, but most existing methods rely on fixed or manually tuned priors, limiting adaptability and stability across diverse conditions. We introduce a structure-guided adaptive total variation (SG-ATV) reconstruction framework that enables stable passive NLOS imaging using a conventional color camera. The key innovation is a structure-guided adaptive total variation (TV) formulation that computes spatially varying regularization weights, automatically balancing detail preservation and noise suppression and overcoming long-standing issues of parameter sensitivity and color noise in passive NLOS reconstruction. The framework dynamically adjusts the regularization weights throughout the iterative process, eliminating manual tuning. Structural information extracted from a fast preliminary reconstruction is used to construct a guidance map that robustly steers the subsequent optimization across different scenes and noise conditions. Experiments demonstrate that the proposed method improves reconstruction efficiency by approximately 30 times while maintaining high robustness and generalization. In terms of quality, it increases the peak signal-to-noise ratio (PSNR) from 17.52 dB to 58.92 dB and reduces color deviations (ΔE = 0) from 28.6 to nearly zero, achieving near-perfect color and structural fidelity. Moreover, the proposed prior-guided mechanism provides a scene-independent weighting strategy that can be directly integrated into other TV-based or optimization-driven NLOS reconstruction pipelines, offering broader applicability without additional parameter tuning.