On the use of Vision for the Weighting of GNSS observation: comparisons

Positioning is a critical function in every intelligent vehicle application. Most applied system is a GNSS (Global Navigation Satellite System)-based receiver, cheap and offering a continuous meter-level accuracy. However accuracy strongly depends on the satellite signal reception state: LOS (Line of sight), that is direct, optimal signal or NLOS (Non LOS), i.e received without direct visibilty and after one or more reflections of the signal. Based on previous work, this paper summarizes different weighting schemes applied to mitigate these local effects on GNSS signals and enhance position accuracy in land transport environment. A database collected by ISAE is used for application and comparison of the different schemes. One of them not only relies on GNSS signals but also on satellite state identification thanks to the use of a fisheye camera. This state allows us to deweight degraded measurements without excluding them, in order to keep availability. The paper shows that considering this additional information allows the WLS (Weighting Least Square) to significantly increase accuracy in every types of environments. If state of the art weighting schemes already improve accuracy by 65 and 77% over ordinary least squares in the 2D plan, our weight further improves accuracy by 40% compared to the classical elevation-based weight.