【摘要】 针对海底管道受水中污垢、海水腐蚀、运行磨损等影响而产生缺陷和损伤问题，设计了一种基于光视觉技术的自主巡检系统并制作机器人样机进行实验验证；应用光视觉技术，设计运动控制器，实现以前端摄像头为运动圆心的寻迹偏角和位移偏差校正，保证了摄像头视野的全程有效性；基于YOLOv4-tiny模型，提出了一种异常综合评估算法，以动态评估运动过程的图像数据，有效避免了由于模型精度有限造成的误识别、数据利用不充分、同步性差等问题；经连续多次水下实验，表明自主巡检系统能够在水平方向、竖直方向准确地沿水下管道进行循迹运动，运动准确率达100%;能够准确识别水下管道吸附物的位置和形状，识别准确度达到96%。更多还原

【Abstract】 In order to solve the problems of defects and damage in underwater pipelines caused by dirt in water, corrosion of seawater, wear and tear in operation, an autonomous inspection system based on optical vision technology is designed, and the experimental verification is conducted by making a robot prototype. In order to ensure the full effectiveness of the camera field of view, a motion controller is designed by using the light vision technology. During the operation of the controller, the front camera is used as the center of the motion circle to realize the tracking deviation angle and displacement deviation correction. An anomaly synthesis evaluation algorithm based on the YOLOV4-Tiny model is proposed to dynamically evaluate the image data of the motion process which effectively avoids the problems of misidentification, insufficient data utilization and poor synchronization caused by the limited accuracy of the model. Many continuous underwater experiments show that the autonomous inspection system can track the underwater pipeline accurately in the horizontal and vertical directions, and the motion accuracy is by 100%; it can accurately identify the position and shape of the adsorbates on the underwater pipeline, and the recognition accuracy is up to 96%.更多还原