Adaptive windowing for photon-efficient non-line-of-sight imaging under high ambient light

Non-line-of-sight (NLOS) imaging aims to recover the shape of hidden objects and has attracted increasing interest in diverse applications. However, practical NLOS systems often suffer from extremely weak return signals buried in strong ambient light, severely limiting imaging performance. Here, we present adaptive windowing non-line-of-sight (AW-NLOS) imaging, which enables robust reconstruction under low signal-to-background ratio (SBR) conditions. Guided by probabilistic models of signal and noise detection arrivals, weak signal detections are amplified via spatio-temporal clustering of correlated pixel blocks, facilitating accurate window size estimation through matched filtering. Per-pixel adaptive short-duration range windows are subsequently applied to effectively suppress overwhelming background detections. Further integrated with total variation (TV) regularization, the proposed method enables high-quality reconstruction even under extreme background conditions, achieving SBR as low as 2.12. Experimental results show that AW-NLOS achieves reliable reconstruction even when photon detection efficiency is severely limited, with signal flux reduced to only 0.02 photons per pixel. Moreover, under a reduced laser power of 42 mW, it improves the structural similarity index (SSIM) of reconstructed objects by approximately 0.5 compared to conventional methods. This work establishes a novel paradigm for precise reconstruction in extreme photon-starved and high-noise NLOS scenarios.