A linear computationally efficient Kalman filter for robust attitude estimation from horizon measurements and GNSS observations. Issue 2 (13th January 2020)

• Record Type: Journal Article
• Authors: Liu, Changhua; Qian, Jide; Wang, Zuocai; Wu, Jin
• Is Part Of: Sensor review. Volume 40:Issue 2(2020)
• Page Start: 153
• Page End: 165
• Publication Date: 2020-01-13
• Subjects: Kalman filter -- Attitude estimation -- GNSS receiver -- Horizon sensors -- Inertial navigation -- Estimation; Sensor systems -- Periodicals; Detectors -- Industrial applications -- Periodicals; Engineering instruments -- Periodicals; 681.2
• DOI: 10.1108/SR-07-2019-0186
• Languages: English
• ISSNs: 0260-2288

Abstract

Abstract : Purpose: For fixed-wing micro air vehicles, the attitude determination is usually produced by the horizon/Global Navigation Satellite System (GNSS) in which the GNSS provides yaw estimates, while roll and pitch are computed using horizon sensors. However, the attitude determination has been independently obtained from the two sensors, which will result in insufficient usage of data. Also, when implementing attitude determination algorithms on embedded platforms, the computational resources are highly restricted. This paper aims to propose a computationally efficient linear Kalman filter to solve the problem. Design/methodology/approach: The observation model is in the form of a least-square optimization composed by GNSS and horizontal measurements. Analytical quaternion solution along with its covariance is derived to significantly speed up on-chip computation. Findings: The reconstructed attitude from Horizon/GNSS is integrated with quaternion kinematic equation from gyroscopic data that builds up a fast linear Kalman filter. The proposed filter does not involve coupling effects presented in existing works and will be more robust encountering bad GNSS measurements. Originality/value: Electronic systems are designed on a real-world fixed-wing plane. Experiments are conducted on this platform that show comparisons on the accuracy and computation execution time of the proposed method and existing representatives. The results indicate that the proposed algorithm is accurate and much faster computation speed in studied scenarios.