Enhanced Visual Pipeline Defect Detection Using Partial Convolution Image Restoration

In urban construction, the defect detection and repair work of drainage pipe-lines are very important, and visual defect detection on pipeline inner surface has become a hot research issue in the application of computer vision and pipeline robot. However, it is still difficult to detect the defects automatically because it is limited by low-quality video and different models of robots. At present, most pipeline detection methods are implemented by pipeline robots equipped with high-definition cameras and manual recognition to find defects frame by frame. To cope with these issues, this paper proposes a method of visual pipeline defect detection enhanced by image restoration. In which, the image restoration using partial convolution is firstly proposed to impair the image that is locally occluded by the haulage rope, then the fast detection using enhanced image data is proposed for pipeline defects. By analyzing the influence of the restoration on the defect detection, the experiment results show that our method has produced significantly improved detection performance by the partial image restoration and it is an efficient method for the application of pipeline robot.