Improvement of Pedestrian Positioning Precision by Using Spatial Correlation of Multipath Error

Pedestrian-to-vehicle communication, which delivers pedestrian position to vehicles to enable real-time estimation of pedestrian-vehicle distance even in the absence of line-of-sight path, has attracted much attention recently. This heavily relies on GPS, whose positioning performance, however, may be greatly degraded in urban canyons due to the influence of multipath error. In this paper, we first investigate the spatial correlation of multipath error. Then, we propose to estimate the multipath error at a pedestrian by (a) using a regression model and (b) Ieveraging the multipath errors at nearby points, which may be, e.g., provided by vehicles that happen to be there. Finally, the pseudo-ranges, corrected by removing the estimated multipath errors, are used to compute an accurate pedestrian position. The proposed method is evaluated with ray tracing simulation and 3D map. Compared with single point positioning without dealing with multipath error, the proposed method greatly reduces pedestrian positioning error by almost one order of magnitude, to 2.2m in the urban areas of Tokyo.