Title: Vehicular Lane-Level Positioning using Low-Cost Map-Aided GNSS/MEMS IMU Sensors Integration
Author(s): Mohamed M. Atia, Allaa Hilal
Published in: Proceedings of the 2018 International Technical Meeting of The Institute of Navigation
January 29 - 1, 2018
Hyatt Regency Reston
Reston, Virginia
Pages: 483 - 494
Cite this article: Atia, Mohamed M., Hilal, Allaa, "Vehicular Lane-Level Positioning using Low-Cost Map-Aided GNSS/MEMS IMU Sensors Integration," Proceedings of the 2018 International Technical Meeting of The Institute of Navigation, Reston, Virginia, January 2018, pp. 483-494.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In
Abstract: Intelligent vehicles require robust, accurate, and efficient real-time lane-level positioning to support new emerging technologies such as autonomous driving, high definition (HD) maps, and advanced driver assistance systems (ADAS). While differential GNSS (DGNSS), Real-time Kinematics (RTK) and other augmented GNSS technologies can provide few centimeter accuracy, the infrastructure and extra computation lead to expensive sensors/systems. This challenge motivated researchers to explore alternative or aiding methods to achieve the same level of accuracy at lower cost. The paper discusses the feasibility of a lowcost lane-level positioning system that fuses measurements from standard GNSS technology, MEMS IMU, and digital road networks without the need for explicit spatial storage of lanes. The system uses a multi-phase approach where a sensor fusion engine is used to integrate GNSS and IMU using extended Kalman Filter (EKF). The second phase is a two-stage HMM-based map-matching algorithm. In the first stage, HMM Bayesian network is built using road-level information only modelling road geometry, map topology, and vehicle motion constraints. HMM network is then decoded by a variable-window Viterbi algorithm. In this implementation, Viterbi algorithm is applied on overlapped variable length windows where size is adaptively adjusted according to the vehicle speed. In the second stage, an adaptive least-squares regression algorithm is used along with lane information to determine which lane the vehicle is moving on. The estimated lane-level position is then applied as a correction to the estimation filter. The system handles GNSS temporary degradation by automatically detecting lane changes using corrected inertial sensor measurements. The system was verified on a 65-km test trajectory showing approximately 97% success rate.