关键词: 月面环境, 深度感知, 立体匹配, 自监督学习, 三维渲染

Abstract: Stereo vision has emerged as a key technical approach for perceiving 3D terrain in lunar rover missions, owing to its advantages of low cost and high reliability. In recent years, deep learning-based stereo matching methods have become the mainstream solution for achieving high-precision stereo vision. However, limited by data acquisition constraints, there is currently a lack of public stereo matching datasets tailored to the lunar surface environment. This severely restricts the training and fine-tuning of deep learning models in lunar scenarios, thereby impairing their adaptability to complex lunar terrain. To address this issue, we propose Render3D, a 3D rendering-based self-supervised learning method for stereo matching. The proposed method requires only monocular panoramic lunar images as input. By integrating the high-fidelity surface rendering capability of 2D Gaussian Splatting (2DGS) with the accurate geometric reconstruction capability of Neural Radiance Fields (NeRF), it generates high-quality pseudo-annotated training samples. These samples guide the fine-tuning of the stereo matching model to adapt to the lunar environment. Experiments conducted in both simulated lunar environment and real physical scenario demonstrate that the stereo matching model fine-tuned using the self-supervised Render3D method significantly outperforms other methods in terms of accuracy. For lunar surface scenarios, the proposed method shows clear superiority over existing self-supervised learning approaches, particularly in challenging conditions such as textureless regions and areas with high shadows, where the matching error is reduced by approximately 50% compared to baseline methods, achieving state-of-the-art (SOTA) performance. Experimental results fully demonstrate that the Render3D method can effectively alleviate the constraint of scarce labeled data for lunar scenes, thus significantly improving the robustness and generalization ability of stereo matching models in the lunar environment.

Key words: lunar surface, depth perception, stereo matching, self-supervised learning, 3D rendering