3D peak based long range rover localization

The dead-reckoning techniques such as visual and wheel odometry for Rover localization is proved to be effective on short to medium-range traverses. However the accumulated error would lead to a degraded performance in a long range travel. In this paper, a global localization algorithm is proposed to deal with the long-range rover localization in lunar/planetary environment. This algorithm is designed to localize a rover by matching 3D peak features extracted from rover imagery with their correspondences extracted from simulated imagery based on digital elevation map. We present a method to represent peak's 3D shape by a multivariate distribution, and match it with other peaks using hotelling's t-square test. After matching, Bayesian network is applied to estimate the position of the rover using matched corresponding peaks. Devon island dataset is used for testing and evaluating the proposed method, as well as comparing it with VIPER. Test results show that proposed method provides an accurate localization in an absolute frame of reference, and achieves a better result than VIPER system.