Design and Experimental Validation of Pipeline Defect Detection in Low-Illumination Environments Based on Bionic Visual Perception

Detecting internal defects in narrow and curved pipelines remains a significant challenge, due to the difficulty of achieving reliable defect perception under low-light conditions and generating collision-free motion trajectories. To address these challenges, this article proposes an event-aware ES-YOLO framework, and develops a pipeline defect inspection experimental environment that utilizes a hyper-redundant manipulator (HRM) to insert an event camera into the pipeline in a collision-free manner for defect inspection. First, to address the lack of datasets for event-based pipeline inspection, the ES-YOLO framework is proposed. This framework converts RGB data into an event dataset, N-neudet, which is subsequently used to train and evaluate the detection model. Concurrently, comparative experiments are conducted on steel and acrylic pipelines under three different illumination conditions. The experimental results demonstrate that, under low-light conditions, the event-based detection model significantly outperforms the RGB detection model in defect recognition rates for both types of pipelines. Second, a pipeline defect detection physical system is developed, integrating a visual perception module based on the ES-YOLO framework and a control module for the snake-like HRM. The system controls the HRM using a combination of Nonlinear Model Predictive Control (NMPC) and the Serpentine Crawling Algorithm (SCA), enabling the event camera to perform collision-free inspection within the pipeline. Finally, extensive pipeline insertion experiments are conducted to validate the feasibility and effectiveness of the proposed framework. The results demonstrate that the framework can effectively identify steel pipeline defects in a 2 Lux low-light environment, achieving a detection accuracy of 84%.