A-contrario approach for outlier detection in GNSS positioning

The localization, that allows to precisely know the orientation and position of a system in the environment, is still challenging in urban environments due to satellite occlusion. This phenomenon reduces the data redundancy and provides bad satellite geometry and reflected signals, called multipaths, that distort the measurements and make them erroneous. The RAIM method, integrated in most GPS receivers, deals with one outlier per instant. This assumption is not longer valid in urban areas where more than one outlier have to be considered. This paper proposes a new approach for the detection and exclusion of the outliers in the GPS pseudo-distance observations. From two assumed models representing the distribution of inconsistent data (naive models), two criteria are proposed to partition the dataset between inliers and outliers. Two algorithms implementing these criteria are presented and evaluated on two datasets respectively acquired in an open environment and in an urban environment. The results are compared with methods classically used for such problems. We show that the proposed outliers detection algorithms improve the estimation of the receiver location and, in the presence of 30% or more outliers, outperform the classical approaches.