基于光照不变性模型的行星探测视觉导航定位方法研究

Extraterrestrial planetary exploration is important to the answers to many fundamental questions including the formation of the universe, the evolution of life and the origin of the Earth. As one of the key technologies in planetary exploration, robust and accurate navigation system ensure the safety and precise of the scientific tasks. The autonomous real-time and active navigation of planetary vehicles and robots is an inevitable trend in the development of deep space exploration. Since no positioning system such as GNSS is established on extraterrestrial planets, vision navigation based on optical images becomes a research hotpot in the filed of autonomous navigation for planetary exploration. The key technology in vision-based navigation is to achieve the real-time and automatic image matching. This task, however, is not always achievable, because there are many featureless areas on planetary surface and the topography information shown on images are largely depend on the illumination condition. To cope with the issues, this project will investigate the mathematical relationship between the planetary topography, illumination and image phase information, and then try to propose an illumination invariant model for planetary based on the previous work. Based on the theoretical model, this project will improve PC (Phase Correlation) based image matching method for feature points tracking in frame-frame matching and image registration in frame geo-referencing. Meanwhile, the project will introduce LBA(Local Bundle Adjustment) and matching error propagation theory in SLAM to achieve robust and accurate vision-based navigation even in featureless areas under considerable illumination variation condition. Through establishment of illumination invariance theory model and the propose of robust and real-time matching algorithms, we strive to promote our country’s research level of vision-based navigation for planetary exploration to keep pace with international counterparts.

行星探测是人类了解宇宙、探索生命起源及地球演化的重要手段，高性能精确自主导航是实现行星探测任务的基础。然而地外行星上并未建立起基于卫星的定位系统GNSS，基于图像的视觉导航逐渐成为行星导航领域的研究热点。视觉导航的核心是快速、准确的图像匹配技术，而行星地貌中存在大量重复性纹理区域，且图像信息受光照影响较大，制约了行星图像自动匹配的实现。针对这一问题，本项目将从行星光照模型切入，探索并揭示行星图像纹理、光照条件与地形地貌三者在频域相位空间中的内在联系，建立光照不变性理论模型，提出连续帧图像跟踪匹配算法和探测器图像与遥感影像整体配准算法，引入SLAM技术中局部光束法平差及匹配误差传播理论，实现在行星纹理重复区域和光照变化条件下探测器自主实时精准导航定位。力争通过光照不变性理论模型的建立和高鲁棒性实时匹配算法的提出，推动我国在深空探测视觉导航方面的研究与国际保持同步。

高性能精确自主导航是实现行星探测任务的基础，而行星地貌中存在大量重复性纹理区域，且图像信息受光照影响较大，制约了行星探测视觉自主导航的实现。针对以上问题，本项目主要建立了基于频域相位相关的地外行星视觉导航光照不变性理论模型，开展了连续帧图像快速配准及探测器影像与遥感影像鲁棒性配准算法研究，基于局部平差方法的相对定位与绝对定位联合导航算法及研究，实现在行星纹理重复区域和光照变化条件下探测器自主实时精准导航定位。本项目建立了首个火星探测卫星无人机联合视觉导航数据集，包含4万余张火星视觉探测模拟仿真图像数据及9种地形对应地面表面模型（DEM）数据，逐步建立并完善了以频域相位互相关功率谱为核心的频域特征匹配理论体系，为行星导航奠定理论基础，将AI深度学习算法与传统视觉导航定位相互融合，搭建行星实时三维场景重建与导航定位软硬件集成原型系统，为实现未来行星探测提供可行性方案。