Absolute Localization of Underwater Pipeline Defects Based on Multimodal Information Fusion and Active Perception

Abstract

With the rapid pace of urbanization, maintaining underground drainage pipelines has become increasingly challenging, particularly in defect localization within complex pipeline terrains, where accuracy and stability are paramount. Existing pipeline robot localization methods often struggle under conditions such as unruly cable deformation and moist, gravelly, muddy, and sandy terrains, which reduce localization accuracy. This paper presents a defect localization framework for a pipeline robot integrating multimodal information fusion and active perception strategies. It mitigates measurement errors caused by cable weight and flexible deformation through an error compensation model that utilizes single-line LiDAR. The localization accuracy in complex environments is significantly improved by dynamically fusing cable and wheel odometry data based on our proposed slip ratio and state estimators. Moreover, an enhanced monocular distance measurement algorithm and an active perception strategy are employed to achieve precise defect localization. Experimental results demonstrate that the root mean square error of the localization in a 32-meter-long urban drainage pipeline is 0.027 m, outperforming traditional methods, with an average defect localization error below 0.03 m. This study contributes to accurate pipeline robot localization in harsh application scenarios.