Passive non-line-of-sight pedestrian imaging based on light transport matrix decomposition

Passive non-line-of-sight (NLOS) imaging extends the observer's perceptual range, offering promising applications in fields such as autonomous driving and emergency rescue. However, mainstream passive NLOS imaging methods are generally limited to distances of less than 6 meters and often rely on a single simulated light transmission matrix, overlooking brightness and reflection characteristics. To overcome these limitations, we employ an infrared camera to capture relayed signals from long distances and introduce the NLOS Pedestrian Imaging Algorithm based on Light Transport Matrix Decomposition (NLOS-LTMD). This innovative algorithm simulates the light transport matrix in two parts: an illumination transport matrix for reconstructing brightness and a reflection transport matrix for enhancing contours. In addition, we propose an Illumination-Oriented Transformer (IO-Transformer) that utilizes threshold segmentation to identify the most informative regions based on the intensity of illumination information addressing the issue of introducing noise and low-SNR regions during contextual modeling, which can degrade imaging quality. Quantitative and qualitative experiments demonstrate that NLOS-LTMD markedly surpasses existing methods in both robustness and reconstruction quality.