Multi-sensor Fusion Odometer Method for UUV

Odometry, as a critical component in Simultaneous Localization and Mapping (SLAM) systems, provides continuous pose variation information whose accuracy directly impacts the overall performance of SLAM. However, single sensor based odometry methods face challenges including initialization difficulties and trajectory drift in specific operational environments. The complementary fusion of multi-sensor data demonstrates enhanced adaptability to environments with varying complexity while ensuring system robustness. This paper proposes a multi-sensor fused odometry method featuring two technical innovations: In data processing, an improved Alpha-Shapes algorithm addresses sonar image noise by treating sonar images as discrete point sets and extracting boundary points for effective denoising. Regarding data fusion, a sub-pixel level phase correlation method enables inter-frame linear velocity estimation from acoustic images, while a visual-inertial fused odometry algorithm is developed based on the Multi-State Constraint Kalman Filter (MSCKF) framework. Comprehensive validation through pool experiments and under-ice field tests confirms the proposed method's robustness in practical underwater.environments.