Research On Welding Seam Line Detection Algorithm Based On Depth Vision

Non-destructive testing technology can effectively ensure the safety and reliability of engineering projects.Ultrasonic non-destructive testing(Ultrasonic NDT)can detect welding quality by manual handheld devices without damaging the test object and affecting the performance of the weld.However,artificial ultrasonic testing is costly and inefficient,and cannot be performed in some special environments.In order to overcome the various disadvantages brought by manual operation,an automatic non-destructive testing method using a robot to carry an ultrasonic testing equipment to work has emerged.The robot structure studied in this paper is an electromagnetically driven wall-climbing robot platform equipped with a navigation system and an ultrasonic non-destructive testing system.The key to the technology is to obtain the relevant motion parameters responsible for robot navigation,which is the focus of this paper.At present,the commonly used welding path detection methods are mainly 2D image analysis methods based on computer vision principles,such as structured light-based navigation methods,neural network-based visual navigation methods,and semi-automatic navigation methods using artificial sticking.After long-term development,these methods have gradually stabilized their performance and gradually enhanced their robustness.However,due to its nature,2D images inevitably have higher requirements for the quality of the images.Even with a highly reliable structured light solution,it may not be a good solution when encountering problems such as oil stains,strong light,and shadows.Based on the navigation requirements of automatic non-destructive testing robots,this paper innovatively proposes a three-dimensional visual navigation analysis method based on three-dimensional reconstruction methods.Due to the addition of three-dimensional point cloud information,compared with the method using only two-dimensional images,it adds more Many on-site information and different detection principles avoid the interference problems such as light,rust,and oil pollution that are difficult to handle with the two-dimensional image analysis method of welding seams.At the same time,the processing speed and accuracy are correspondingly improved,which greatly improves its robustness.The hardware solution uses Intel Realsense series image sensors,which can simultaneously capture two-dimensional image information and three-dimensional point cloud information,and determine the weld boundary by analyzing the captured information.The overall logic of the detection scheme in this article is divided into four parts: The first part is information acquisition and preprocessing.The main content is image acquisition,cropping,filtering,accuracy testing,depth map conversion,etc.The second part is the welding seam segmentation.This paper attempts to use a variety of algorithms to accurately segment the weld seam and base metal parts from the point cloud in order to facilitate the boundary extraction of subsequent algorithms.The third part is to extract feature points from the segmented information,and use the feature points to fit the welds that meet the requirements.At the same time,the RGB information is also analyzed,and the results are merged with the point cloud analysis results to extract the boundary to improve the robustness of the system.The fourth part is the statistical experiment results,precision analysis and introduction of equipment related attributes.After the system is packaged,field verification is performed,and comprehensive tests are performed for different welds.A total of 15 test experiments are performed,and the two pose parameters of distance d and angle θ are counted.The results show that the accuracy of d is 0.53 mm,the accuracy of θ is ± 0.84 °,and the error meets the requirements of working conditions.