The Research Of Laser Visual V Seam Image Recognition

Seam sensor based on laser vision has the characteristics of obtaining a large amountof image information, high reliability, wide use etc. In the actual work environment, dueto welding working conditions can not be very satisfactory, and the arc sensor is near tothe weld acr, there are arc strong, high temperature and radiation, electromagneticinterference, dust flying and other negative factors. Therefore, the images acquired bysensor can not be the ideal image, it needs some processing steps to improve the imagequality for the observation and evaluation, extracting useful information parameters ofthe control system information parameters.First the structure and hardware design of image recognition system selection havebeen determined, using DSP digital image processor can quickly batch image processing,image acquisition, image batch processing, data storage and other processes. Byestablishing a mathematical model of laser structured light sensor, it achieves theconversion of three-dimensional coordinate space and two-dimensional coordinates of theimage. Optimizing the analysis to mathematical model for structured light sensordetermines the installation position of the camera, a laser light source, which optimizesthe structural parameters of the sensor.By analyizing the model of the camera image, calibrating the image recognitionsystem and using the method of a linear least squares fitting, the relation among pixel sizeand actual size is obtained.According to the characteristics of V-groove weld, it developed a seam imageprocessing algorithm. By the process on image filtering, thresholding segmentation, edgedetection, ossification, linear extraction, the information of the weld in the process isextracted. First, it carrys on the weld image processing and image smoothing thresholdsegmentation, compares the effect of various filtering algorithms eventually determinesmedian filtering algorithm. Second, canny edge detection algorithm can clearly detect theboundary portion of the laser light bands by edge detection. Third, the results on edgedetection are ossificated, extracting a single pixel of centerline on the laser light with.Finally centerline is line-fitted by using hough algorithm.By laser light with a single-pixel line after fitting, the pixel coordinates of V-groovewith two shoulders and one at the bottom are extracted. It gets the V-groove pixel widthand depth and derives the actual size information by calibrating the relationship.