一种管道蛇形机器人的裂缝视频检测系统

为实现蛇形机器人在管道内部快速准确的识别管道内壁裂缝,基于一种改进YOLOv3算法为管道蛇形机器人设计了快速检测管道裂缝的系统。系统搭载了500万像素相机以及用于辅助标定的两个激光发生器。此系统通过摄像机采集管道内部视频信息,使用改进YOLOv3算法对视频进行检测,若识别出裂缝则输出当前图像。之后结合激光标定和边缘检测算法得到当前裂缝的物理信息。改进YOLOv3算法使用k-means++算法对裂缝数据集进行聚类,得到最佳先验框,并使用距离交并比代替交并比作为损失函数,以提高精度和速度。实验表明,改进YOLOv3算法平均精度为87.23%,与原始YOLOv3算法相比提高了5.88%；同时基于激光标定算法的图像处理得到的裂缝物理信息与实际信息误差在5%以内,可以用于实际工程。

A Crack Video Detection System of Pipeline Snake Shaped Robot

In order to realize the snake-like robot to quickly and accurately identify the cracks in the inner wall of the pipeline, a system for rapid detection of pipe cracks is designed for the pipeline snake shaped robot based on an improved YOLOv3 algorithm. The system is equipped with a 5-megapixel camera and two laser generators for auxiliary calibration. The system collects video information inside the pipeline through the camera, detects the video using the improved YOLOv3 algorithm, and outputs the current image if a crack is identified. After that, the laser calibration and edge detection algorithms are combined to obtain the physical information of the current cracks. The improved YOLOv3 algorithm uses the k-means++ algorithm to cluster crack datasets to obtain the best prior box, and uses the distance cross-to-merge ratio instead of the cross-merger ratio as a loss function to improve accuracy and speed. Experiments show that the average accuracy of the improved YOOlOv3 algorithm is 87.23%, which is 5.88% higher than that of the original YOLOv3 algorithm. At the same time, the error between the physical information and the actual information of the crack obtained by the image processing based on the laser calibration algorithm is within 5%, which can be used for practical engineering.