Effect of Coordinate Transformation on Error Covariance and a Method to Improve Accuracy of Self-localization with a Camera Fixed to a Movable Part

Abstract

A method of improving accuracy of UKF-based self-localization through the proper evaluation of error covariance matrices and its effect over improvement of self-localization of autonomous vehicles are suggested. First, error covariance matrices of spatial coordinate transformations defined as 1) an inverse of a coordinate transformation whose error covariance matrix is known, and 2) a synthesis of two coordinate transformations whose error covariance matrices are known, are derived. Second, a vehicle with a camera attached to a movable part favorable to landmark detection is presented. We demonstrate a case in which the accuracy of landmark-based self-localization of a vehicle is improved by using a camera fixed to a movable part that allows better tracking of landmarks, such as a steering rod of a micro-electric vehicle, if the error covariance matrices to be fed to UKF reflect the mechanical noise properly.