3D vehicle positioning with smartphone-based hybrid techniques in urban areas

Abstract

With the increasing demand for vehicle navigation in urban areas, accurate and reliable vehicle navigation is attracting extensive attention in Intelligent Transport Systems (ITS). Due to their widespread adoption and various benefits, smartphones become the first choice of road users for vehicle navigation. Smartphones can provide a global covered positioning accuracy of 5-10 m in open areas with embedded Global Navigation Satellite System (GNSS) modules. However, with the development of cities, grade separations, such as overpasses and tunnels, are increasingly used in road design. The GNSS positioning performance can be severely degraded to more than 50 m especially in the vertical direction because of satellite signal blockage and severe multipath effects in dense urban areas. An embedded Inertial Measurement Unit (IMU) is usually integrated with GNSS to smooth its trajectories and provide continuous locations. However, due to its low quality, IMU cannot maintain accurate positioning for extended periods. Additionally, the integrated solutions are susceptible to errors from GNSS positioning results since accurately estimating GNSS errors is challenging in urban environments. The positioning accuracy of the integrated system is not sufficient for the requirements of urban 3D navigation. There is a pressing need to enhance smartphone positioning performance in urban areas. In this study, to provide accurate altitude and road layer for vehicle navigation in multilayer road networks, the barometer is introduced into the GNSS/IMU integrated system with the adaptive interpolated Mean Sea Level (MSL) pressure values, the GNSS measurements, and Digital Surface Model (DSM). After that, to improve the localization performance of the integrated system in long-period GNSS outages, a deep odometry network, named DeepOdo, is designed for estimating the forward velocity of the vehicle. Finally, to calibrate the position and direction of IMU results and estimate the accuracy of GNSS in urban canyon areas, the Map-Matching feedback (MMF) algorithm is utilized with the digital map. Extensive experiments have been done in dense urban areas of Hong Kong to assess the capability of the proposed methodology. The experimental results demonstrate that even in urban canyons, the integrated positioning algorithm proposed in this study yields average errors of less than 10 m (9.7 m) in the horizontal direction and less than 1.5 m (1.3 m) in the vertical direction. Furthermore, in GNSS-denied areas, the utilization of DeepOdo for forward velocity estimation and barometer data led to a substantial reduction of 78% and 99% in horizontal and vertical positioning errors, respectively, when compared to relying solely on the IMU. These experimental findings attest to the reliability and accuracy of the proposed method, making it a promising solution for enhancing navigation experiences for smartphone users in complex urban environments.