Dynamic Mode-Adaptive Optical-Inertial Fusion for Robust Pose Estimation

Pose estimation algorithms based on optical-inertial fusion encounter significant challenges, such as the loss of optical pose measurements due to occlusion, outliers caused by magnetic interference and motion-induced accelerations, as well as significant residuals in gyro bias calibration when deployed on oscillating platforms. To address these issues, a dynamic mode-adaptive optical-inertial fusion pose estimation algorithm is proposed. To calibrate gyro bias simultaneously during motion, a Kalman filter model integrating gravity, magnetic field, and optical measurements is designed. Furthermore, a statistical similarity measure method is employed to enhance robustness against magnetic distortions and motion-induced outlier interference. Experimental results demonstrate that the proposed method achieves significant improvements over traditional methods on multiple scenarios and sensor datasets, while effectively calibrating gyro bias residuals in dynamic conditions, highlighting its robustness and adaptability.