Research On The Visual-Inertial Localization Techniques Of The Lunar Rover In Low Texture Environment

China's lunar exploration project clearly pointed out that it is necessary to establish the Moon Base and develop space industry in the future.The construction of a large-scale Moon Base requires that the detection area of the lunar rover be changed from a small-scale detection near the landing area to a long-range detection.Restricted by the insufficient resolution of the existing lunar elevation map,it is difficult to guarantee the high-precision motion planning capability of the lunar rover.Therefore,strict requirements are placed on the high-precision positioning of the lunar rover in a large-scale unknown complex environment.With the expansion of the movement range of the lunar rover,the terrain environment passing by is complex and diverse.Compared with the crater environment with higher texture,the perception system of the mobile module in the low texture environment has the following difficulties:(1)The lunar surface can be extracted few visual feature points,which leads to low data association;(2)The feature point descriptors have high similarity and serious mismatches,which directly lead to insufficient positioning accuracy.Therefore,this paper proposes a high-precision visual-inertial positioning method to improve the positioning accuracy of the lunar rover in a low-texture environment,so as to ensure the smooth development of the detection task.Aiming at the difficulty of insufficient feature point extraction in low-texture environment,this paper proposes to add line features to the positioning system.First,the Line Segment Detector(LSD)is used to extract line features in a low-texture environment.Second,the binary Line Band Descriptor(LBD)is generated,nearest neighbor matching is performed,and the matching outliers are deleted using geometric constraints.Finally,using the Plücker coordinates to parametrically represent the line features,and establish the observation model of the line features.To deal with the problem of serious mismatch caused by high feature point similarity in low texture environment,this paper designs a feature point matching algorithm based on motion smoothness,and applies it to the feature point matching module of the positioning system.First,this paper uses inertial information to limit the feature point matching area,and then remove outliers based on an improved grid motion statistical algorithm.Finally,the comparison and verification are carried out on the three data sets of TUM,Devon Island,and KITTI.The test results show that the matching accuracy and matching rate of the feature point matching algorithm based on motion smoothness in the low texture environment are better than the traditional algorithms.This paper designs a visual-inertial positioning system for the lunar rover in a low-texture environment,and establishes a visual-inertial tightly-coupled nonlinear graph optimization model based on the combination of point/line features,and uses sliding window optimization to reduce computational complexity.The positioning system was tested on the Eu Ro C dataset,the CAR dataset,and the NASA lunar elevation map simulation dataset.The experimental results show that the addition of line features and feature point matching algorithms based on motion smoothness in low-texture environments can improve the positioning accuracy.The positioning system proposed in this paper is robust to changes in illumination and is suitable for lunar terrain environments.