Coupling Outlier Detection with Particle Filter for GPS-Based Localization

In constrained environment such as urban areas, vehicle localization, using Global Navigation Satellite System (GNSS) data, is impaired by multipaths and Non Line Of Sight effects. These latter induce erroneous measurements that prevent the accurate estimation of the receiver position. In this paper, we propose to detect and discard these outliers in order to improve the localization. We propose a new algorithm to detect the outliers either in the pseudo-range observations and/or in the Doppler observations. Based on a model representing the distribution of inconsistent data, a criterion is proposed to partition the data between inliers and outliers. This criterion is coupled with a particle filter used for localization. Coupling is carried out in terms of selection of the data (pseudo-range and/or Doppler measurements) used in the particle filter. Besides, considering only GNSS data, the estimation of the instantaneous velocity (required for position prediction) is derived from Doppler measurements. An experiment with a GPS receiver on board a vehicle has been carried out on constrained environments, forest and urban canyon. Results prove that the proposed algorithm is both robust to the presence of 'outliers' in the raw data (thanks to the a-contrario detection step) and robust to non Gaussian noise and satellite blockage (thanks to the particle filter prediction step). Specifically, localization errors decrease of about 1/3 relative to the GPS receiver solution (also using pseudo-range and Doppler measurements).