Hierarchical outlier identification-based novel magnetic sensor correction and magnetic/inertial sensor fusion scheme for roll angle estimation of spinning projectiles

Low-cost magnetic and inertial sensors are widely used for attitude estimation, yet their performance is limited by magnetic interference and inertial sensor drift. To address these challenges, this paper presents a hybrid correction framework that integrates roll angle measurements from both sensor types for precise estimation. First, we propose a hierarchical outlier detection-based ellipse fitting method to compensate for magnetic measurement errors. Leveraging the inherent data structure, a novel Lightweight Incremental Local Outlier Factor (LILOF) algorithm with a dual-threshold strategy is developed for coarse outlier identification in the outer layer. This is followed by a RANSAC-based refinement step using Sampson distance to eliminate residual anomalies, achieving an optimal balance between computational efficiency and geometric accuracy. Subsequently, a fusion algorithm combines geomagnetic and MEMS gyroscope data via a Sage–Husa fading adaptive Kalman filter (SHFAKF), where the Sampson distance criterion dynamically weights sensor inputs to mitigate their individual limitations, leading to improved accuracy in attitude calculation. A fading factor is further incorporated to enhance stability and prevent filter divergence. Finally, simulations and ground experiments demonstrate our proposed method's superior accuracy and robustness compared to other advanced approaches.