Large-scale fusion of collective, areal vehicle data

In this paper we present a novel approach to the large-scale fusion of collaboratively acquired areal sensor data from a vehicle fleet, like temperatures, illuminations, frictions, traffic densities, signal strengths, air qualities, etc., which are intended for the incorporation into next-generation comfort functions and driver assistance systems. Our algorithm is based on the interpolation via RBFNs, but is extended, so that it fulfills the additional demands of the automotive industry. Therefore, we examine how to tackle large-scale, integrate timestamp-dependent weights, incorporate elliptic basis functions and realize an incremental computation. Finally, we investigate how our incremental approach to interpolation can be further improved by incorporating the Fast Gauss Transform, resulting in a reduction of the computation time by a factor of ten, if a batch insert or update (e. g. in the case of temporal reweighting) has to be applied. Our suggested approach to double-staged interpolation is applied exemplary to friction data, but can be also applied to other kinds of areal data as enumerated prior. Further, we present, that the first stage of the double-staged interpolation can be utilized for the determination of lane geometries, which can be consecutively used to annotate fused areal data with appropriate lane affiliations.