A Study on Correspondence Point Search Algorithm Using Bayesian Estimation and Its Application to a Self-position Estimation for Lunar Explorer

The purpose of this research is to improve the accuracy of the lunar surface high-precision landing technology for small unmanned probes in the SLIM (Smart Lander for Investigating Moon) project [1] of JAXA. As a method for realizing high precision landing technology, matching by image collation using the images taken by the space probe and the lunar map image is being studied. However, shooting in outer space tends to cause positional errors due to low resolution caused by disturbances. Conventionally, position estimation has been performed by matching based on k-NN Matching and Ratio Test. However, the accuracy of the estimated position depends on the selected correspondence points, and no error correction is performed in this method. Therefore, we propose a matching model that minimizes the error of position estimation by using Bayesian estimation. Bayesian estimation is an estimation approach that uses the likelihood calculated from observed data and the prior distribution of parameters. In the proposed method, the estimation problem based on the feature values of local features is formulated as a constrained optimization problem, and a Bayesian model is constructed by designing the likelihood and prior distribution. In this paper, we realize a matching algorithm with the function of conventional self-localization by Bayesian estimation.