Abstract

Accurate vehicle positioning is becoming increasingly important for modern transportation systems, especially with the recent focus on advanced driver assistance systems and autonomous driving. Currently Global Navigation Satellite System (GNSS) is the main source for vehicle positing information. However, GNSS can suffer from reduced accuracy or complete signal loss due to signal blockage or multi-path errors, particularly in urban environments. In systems that rely on accurate positioning, such failures could pose a major safety risk to passengers, other road users and surrounding infrastructure. Therefore, this thesis investigates an alternative solution to estimate vehicle position. By utilizing data from wheel speed sensors and steering angle sensors together with motion models, vehicle trajectories are calculated and evaluated. The study also investigates how motion models can be combined with GNSS measurements using an extended Kalman filter (EKF) with the goal of improving heavy vehicle positioning. In total, five different motion models, three bicycle models of varying complexity, one rear axle model and one front axle model are evaluated using data collected at a test track. An RTK GNSS is used as the ground truth for the experiments.

The results show that motion model selection and implementation has a large impact on the dead reckoning performance. Simpler kinematic models perform better in driving conditions with lower velocity, acceleration and turning. The dynamic bicycle model does not perform as well in less demanding conditions but show good results when tested in more challenging driving conditions. The rear axle model performs accurately in experiments with constant velocity or heading, but becomes less accurate during wide turns and velocity changes. The front axle model performs worse than all other models regardless of the driving conditions. When compared to GNSS, the dead reckoning models achieve higher positioning accuracy for short periods of approximately 10 – 20 seconds, after which the trajectory starts to drift due to accumulated errors. The EKF sensor fusion approach produce a more accurate and robust trajectory estimate compared to standalone dead reckoning. Overall, the results suggest that wheel sensor-based dead reckoning can be combined with GNSS through sensor fusion to enhance heavy vehicle positioning accuracy, or serve as a short-term complementary positioning solution during GNSS outages