Real‐Time LWIR Imaging and Ranging via 4D Light‐Field Metasurface‐Transformer Integrated Framework

Abstract

ABSTRACT Real‐time ranging of thermal targets holds critical importance for defense, surveillance, and autonomous navigation, yet remains fundamentally constrained by the inherently low texture of thermal imagery. A metasurface‐Transformer architecture is presented to enable monocular four‐dimensional (4D) thermal light‐field imaging and subsequent ranging. The optical front end uses an all‐silicon metasurface microlens array directly integrated into a commercial thermal camera module, forming a compact light‐field imager. The computational back end, a thermal disparity transformer (TDiT), resolves sub‐pixel disparities from the texture‐scarce thermal light‐field through dedicated pre‐processing and Transformer‐based inference. The system achieves centimeter‐scale accuracy, with a mean ranging error of 1.87% at 25 ms per frame, enabling real‐time dynamic ranging of moving thermal targets. By combining metasurface‐enabled light‐field capture with high‐precision disparity estimation, this work provides a promising approach for effective, real‐time perception in texture‐scarce thermal environments.