Research On Image Recognition Algorithm For Welding Seam Of Welding Robot Based On Structured Light

Welding manufacturing technology is a very important part of the industrial production process and is an important indicator of the level of industrialisation.The use of industrial welding robots to improve the efficiency of welding production is a trend for the future.The use of machine vision to guide industrial robots in welding has reached a high level of interest.Structured light vision sensors for weld seam positioning offer the advantages of non-contact and fast detection.According to actual research,it has been found that the current vision-based weld seam detection system has disadvantages such as low real-time and insufficient universality,especially due to the smooth surface of the weld material with high reflection effect,making the weld seam features inconspicuous,making the industrial welding robot positioning accuracy insufficient and the welding effect unstable.First,the design and construction of the weld seam image feature recognition and extraction platform and the selection of the main components are completed.Based on the triangulation principle to select the optimal measurement method,based on the line structure light vision design,select the appropriate weld seam positioning sensor,and complete the vision system to the main components of the welding platform selection.The design and construction of the structured light vision platform was completed by selecting a mounting build method to fix the position of the industrial camera in relation to the linear laser generator.Secondly,in order to extract the weld seam image feature information,the pre-processing process is simplified by studying and analysing different image processing methods.A PSPnet network-based weld image processing algorithm is proposed to pre-process the weld image.Using the original images of four typical weld seams collected,the dataset is completed,and the PSPnet network model is trained to complete the weld image segmentation process and extract a clear structured light stripe image.Then,the weld image feature point extraction method was investigated for the weld seam images.By analysing and comparing different image centre extraction algorithms,the Steger algorithm was determined to extract the structured light streak centre line;according to the weld image characteristics of T-shaped weld,V-shaped weld,lap weld and butt weld,the Center Net network was proposed to carry out feature point identification and extraction of four typical weld seams,and comparing the identification and extraction speed of traditional feature point identification algorithms,the The Center Net algorithm used in this paper has a good image processing efficiency while finding the feature points of weld seams.Finally,the experimental system is designed and debugged according to the research content of this paper,and the extraction results of multiple sets of weld seam recognition are compared with the standard points in the validation set.The final experimental results show that the average error in pixel coordinates is 0.877 pixel in X-axis and 0.705 pixel in Y-axis.