A Hybrid State Representation-Based GNSS Filtering Model to Improve Vehicular Positioning Performance

In the era of intelligence, there is a huge demand for precise positioning in the fields such as automatic driving, assisted driving, and vehicle to everything. Due to the continuous and high-precision positioning capability, the global navigation satellite system (GNSS) has been an important foundation for large-scale location applications in open-sky conditions. For land vehicle kinematic navigation, the GNSS positioning method based on Kalman filtering (KF) has also become the primary choice. However, the general state model based on velocity and acceleration in KF lacks mining and utilization of the physical laws of vehicle motion, resulting in the insufficient performance of GNSS KF positioning in urban conditions. Thus, we constructed a hybrid state model, where the vehicle kinematics are classified into nonlinear and quasi-linear motions, and the corresponding velocity constraint (VC) and heading constraint (HC) models are constructed. Then, the reparameterization between VC-model and HC-model is derived through a strict formula, and the hybrid state model can switch adaptively by vehicle motion recognition. Field vehicle test data was collected with a low-cost GNSS receiver and processed by different methods. The test results of different observation conditions and different motions show that the hybrid state model has better positioning performance, and the root-mean-square (rms) errors in the occlusion condition are reduced by 15% compared with only the VC-model. Furthermore, the test results using the public data of the Google Decimeter Challenge project also verify the effectiveness and practicability of the filtering model in vehicular positioning.