Smartphone-based Vehicle Tracking without GPS: Experience and Improvements

Abstract

Nowadays, GPS and other global positioning systems have been widely developed, enabling accurate and convenient outdoor location-based services for vehicles. However, there are still two percents of areas in urban city that cannot be covered by satellites, e.g., underground parking lots, tunnels, and multi-level flyovers. Current positioning methods always rely on inertial dead-reckoning methods, but the performance is seriously affected by the low-quality inertial sensors embedded in crowdsourced smartphones. Based on our series of experiments with thousands of smartphones, we observe that the accuracy of existing inertial dead-reckoning methods is terribly affected by many factors, e.g., arbitrary and unknown placements of smartphones in car, inconstant inertial noises, and the diversity of smartphones and vehicles. In this paper, we explore a novel smartphone-based inertial sequence learning approach to infer vehicle's location in real time. We also propose a customized model refinement mechanism for individual drivers. Extensive experiments on DiDi ride-hailing platform have proved the effectiveness of our solution.