A two-stage network for iToF depth image restoration in foggy environments

摘要

The quality of iToF (indirect-Time-of-Flight) depth images is severely degraded in foggy environments due to strong scattering and absorption effects, resulting in noise and distortion in the depth images. This degradation limits their practical applications in fields such as autonomous driving, robotic navigation, 3D reconstruction, and industrial inspection. To address this issue, this paper proposes a depth image restoration method tailored for foggy environments. A depth dataset under high fog interference was constructed to support the training and evaluation of subsequent models. In terms of algorithm design, a two-stage depth restoration network is proposed. The first stage is an IR (infrared) image deblurring network, which focuses on mitigating the impact of fog on IR images. The second stage is an IR-guided depth restoration network, which combines features from IR and depth images, generates an initial depth map through the backbone network. A Spatial Propagation Network (SPN) is then used to refine the generated depth maps, producing high-quality restored depth images. Experimental results demonstrate that the proposed method achieves significant performance in depth image restoration tasks under foggy conditions. Compared with existing methods, the proposed network achieves higher accuracy and better reconstruction quality in terms of depth. This study provides a novel approach to addressing the quality degradation of depth images in foggy environments and lays a foundation for high-quality depth sensing technologies in practical applications.