Research On Feature Extraction And Matching Based Vision Navigation For Planetary Landing

With the development of space exploration, a novel navigation method needsto be proposed in order to meets the requirements of planetary landing mission. Thispaper focused on the problem of image feature extraction and matching based visionnavigation algorithm. The main content of this paper includes: Clarification oftechnical requirements of feature extraction and matching algorithm for planetarylanding mission; comparison and evaluation of different algorithms for planetarylandng and researches on region feature extraction and matching algorithms.Firstly, we analyzed the powered decent phase of lunar landing, parachuteaided landing phase and powered decent phase of mars landing. Based on theclassical planetary landing trajectory, the technical requriements of featureextraction and matching algorithm for vision navigation in planetary landingmission are proposed.According to the technical requriements proposed in chapter II, five featureextraction algorithms are selected to evalute their application potential in visionnavigation,a new feature matching algorithm based on initial homographyestimation is proposed. After a series of comparison and evaluation, we considerHessian-Affine as the most suitable algorithm for vision navigation’s application inplanetary landing mission.Also the region feature extraction and matching algorithm are reserached inthis paper, a crater detection algorithm based on MSER is proposed in chapter IV,through the pairing of dark and light regions in an image, a vector is generatedconsidering the direction of sunlight to describe the crater. Results from simulationindicate that the correct detection rate of this method can reach above95%, also themethod works well in detecting complex crater. Aiming at landmark basednavigation, we first simplify the crater into a T-model, based on the3D model ofcrater, the area of shadow is extracted to match crater in2D image. Also we analyzethe rock detection algorithm based on K-means, and restore the height of rocks withthe known information of spacecraft’s height. Finally, a semi-physical simulation iscarried out to verify the feature extraction and matching algorithms mentionedbefore.