High-resolution and real-time non-line-of-sight imaging based on spatial correlation

Abstract

Non-line-of-sight (NLOS) imaging technique has broad application prospects in fields such as autonomous driving and robotic vision. Among the existing NLOS imaging methods, the scan-free methods allow rapid data acquisition. However, the imaging resolution is limited by the system's temporal jitter and the number of pixels in the array detector, and the real-time recovery of complex dynamic scenes is still a major challenge. Here, based on the temporal and spatial broadening of the signal and the evolution characteristics of optical transmission fields, we establish the three-dimensional blur kernel and the forward evolution model under scan-free conditions. Subsequently, leveraging the spatial correlation between adjacent detection region information, we propose a resampling method to obtain high-resolution data information with fine collection. Combining this with three-dimensional blur kernel modeling makes high-resolution imaging of hidden targets realized through model inversion. Our method improves the lateral resolution of the imaging results and enables the reconstruction of dynamically complex scenes. We demonstrate high-resolution NLOS imaging at 5 frames per second for dynamic scenes, providing valuable insights for practical applications of NLOS imaging.