Research On Vision-based Localization Technology Of The Lunar Rover

The lunar rover which is actually a kind of mobile robot developed by our country is used to detect surface of lunar. To complete the complex surface topography detection,artificial intelligence, automatic control, computer technology and other scientific technology is gathered as a whole in this field. How to make this robot on the moon get the precise navigation positioning is what we focus on in this thesis.The detection of the moon’s surface with accurate navigation and positioning is depend on visual positioning. The technology can be improved to achieve grate accuracy of visual navigation and positioning. It is very important to moon exploration program that how to improve the accuracy of visual navigation and positioning.This thesis research the technology about the visual positioning used in our“Yutu” Rover. Accurate matching can be used to match the feature points as the prerequisite. Because, The cylindrical projection of sequential images in the current station contains a larger overlapped region with that in the last station than a single image, Thus, the rover in large-span states can be positioned by constructing bundle adjustment models with the matched feature points and the calibration parameters of binocular cameras. With this method, the surface tour guide with the detector is completed. Through the experiment, during the operations, it proves that the new method proposed in this paper can greatly improve the location accuracy of the lunar rover on the surface of the moon.Depend on the appearance of the traditional matching methods, SIFT is widely used in the detection of surface. But with the use of traditional methods, they often fails in some special conditions. One important reason is that the two moon image have same surface of the landscape painting. Around the same visual description, the feature points may exist with lots of differences. By using these traditional methods only consider the appearance of the similarity between feature points, they always ignore the other useful information in the image, such as the structure of the information between feature points.So as a research point in this thesis, we introduce the graph matching algorithm based on the structure. For the future detection of surface preparation for visual positioning technology development, the idea of introducing the structure characteristics has certain advantages. On the one hand, the introduction of the structure information will also make the feature points matching with no effect on geometric transformation andlighting changes. On the other hand, optimization problem should be proposed by spectral technology effective approximate solution.Graph matching algorithm in this thesis is proposed based on the non-convexity and concavity procedure. On the basis of predecessors’ research, this thesis designs the application of the objective function of similar matrix to non convex gradually concave process to complete the external problems of graph matching points. Not only the undirected graph matching but also the directed graph can be applied at the same time with the efficient robust graph matching. Finally, the experimental results of this paper prove the correctness and effectiveness of the proposed algorithm.