Lunar Soft-landing Vision Navigation And Descending Sequential Imagery Match Research

With the fast development of space science and technology, deep-space probing, which is regarded as an important mark of nation comprehensive science technique level, have re-focused by each country in 21st century. Till now, China has already successfully shot two satellites as: Chang'e-1 and Chang'e-2 . Soft-landing of lunar rover becomes a highlighted issue of next lunar probing project. Compared with the former landing technologies, vision navigation aided landing would bring new and better supporting for self-adaptive, precise landing and barrier evading.Job of this thesis mainly including: systematic plan analyzing and designing of lunar soft-landing vision navigation; then, self-adaptive sequential image matching is discussed, which is a basic requirement of 3D reconstruction of lunar terrain and landing spot decision in the landing vision navigation process. Main content of this thesis are listed as following:1. Characteristics of three independence navigation modes and vision navigation are discussed, and background of the thesis is introduced.2. A plan concerned about lunar soft-landing vision navigation is proposed. The research ietms of this plan is presented, and relative hardware system and software system is designed baiscally. The main task in the three stages of vision navigation is introduced in the thesis.3. Based on the comparison of general characteristic extraction methods, Forstner is selected to extract image characteristic point. After analyzing matching results of different dimensional eigenvector in SIFT, 128-dimension eigenvector is used. Then, a combined method including Forstner and SIFT is formed. Experiment result shows that sequential image characteristic point matching can be finished in fast and high accuracy way.4. Based on random sampling consenus (RANSAC) and least square matching (LSM), affine-transformation model is caculated from SIFT corresponding point-set, and used as structure relation of sequential imagery. With affine-transtormation parameters calculation, wrong corresponding point pairs are filtered effectively. Then, image matching of non-characteristic point is done as: firstly, initial matching position of the given point is calculated through affine-transtormation parameters; secondly, local window's image relative coefficiency calculation is used for corresponding point searching; the third, best corresponding position in sequential image decided with peak coefficiency. Experiment proved that, the match result about non- characteristic points is realized fast and accurately, the averge error of corresponding point is less than 3 pixels.5. A hardware-in-the-loop simulation model is built by 6-degress-of-freedom platform and proportion sand table in lab. In this simulated condition, sequential imagery matching method in the paper is effive, and due experiment speed and reliability is obtained.