基于视觉-惯性-压力融合的水下定位方法

In underwater unstructured environments, robots face difficulties in relying on external base stations for localization. Therefore, autonomous localization using multi-sensor fusion has significant application value in such settings. This paper aimed to address issues such as poor stability in visual localization and substantial drift in inertial navigation within underwater multi-sensor fusion localization and proposed a tightly integrated multi-sensor fusion localization method that combined visual, inertial, and pressure sensors. By utilizing graph optimization techniques for multi-sensor fusion and identifying errors in visual-inertial data based on depth information, the quality of the fused data was enhanced. To address drift and localization loss during the fusion localization process, a depth sensor was employed for weight allocation to provide more detailed system initialization. Additionally, closed-loop detection and relocalization methods were introduced to effectively mitigate drift and localization loss. Experimental validation demonstrates that the proposed fusion localization algorithm improves accuracy by 48.4% compared to visual-inertial fusion localization methods, achieving superior precision and robustness. The actual positioning accuracy can reach the centimetre level.