Real-Time Smoothing Filter Based on Constraints for Backward Filtering

Abstract:

The real-time information of systems has been the subject of extensive studies. Although the state of a system in real time can be estimated using only the Kalman filter, this paper focuses on a method for achieving a more accurate estimation. A smoothing filter (SF) can be utilized to improve the estimation accuracy. To implement this in real time, studies have explored the application of a windowing method to develop a real-time SF (RSF) However, achieving real-time performance remains challenging, as the SF has minimal influence on estimation performance for the current time point at the end of the applied interval. To address these challenges, a real-time constrained SF (RCSF) is proposed in this study, establishing constraints for the backward filter through the definition of prior information. Subsequently, a global positioning system/dead reckoning (GPS/DR) navigation system is implemented. A temporal analysis is conducted to demonstrate the feasibility of real-time implementation of the RCSF. Finally, the RCSF is validated through a comparative analysis with other filters using the KITTI dataset. This study contributes to the advancement of real-time estimation techniques, enhancing the real-time performance and applicability of the conventional RSF, which has limitations in current state estimation.