3-D Measurement and Reconstruction of Space Target Based on RAW Images

Abstract

The 3-D surface measurement and reconstruction of noncooperative targets are critical prerequisites for subsequent complex tasks such as target locking, tracking, rendezvous, docking, and landing. The space environment has a single light source and lacks atmospheric diffuse reflection effects, which makes observation challenging. Moreover, imaging modes that simulate human visual perception convert high dynamic range (HDR) RAW images into more storage-efficient standard red green blue (sRGB) formats, resulting in the loss of significant details. Therefore, neural implicit surface methods that use sRGB images often result in significant errors in such scenarios. To solve the above problems, this article proposes a 3-D surface measurement and reconstruction framework based on RAW images—RAWSurf. First, the HDR information in RAW images is used for supervision to enhance measurement and reconstruction accuracy. Moreover, to mitigate the impact of large magnitude spans of RAW images and low signal-to-noise ratio (SNR) of underexposed areas on reconstruction accuracy, a soft weight coefficient mapping is adopted. Meanwhile, use a progressive sampling (PS) strategy to ensure that the model focuses more on the spatial area. Then, by integrating three different state-of-the-art (SOTA) models with our framework, the average Chamfer distance error was reduced by 74%, the average Hausdorff distance error was reduced by 63%, and the average $F{1}$ -score (%) was increased by 11.8. The code is publicly available at https://github.com/liyuandong145619/rawsurf.