Welding seam detection and feature point extraction for robotic arc welding using laser-vision

Abstract

A welding seam detection and feature point extraction system for robotic arc welding is proposed in this paper. The system consists of a CCD camera, a line structure laser, a narrowband optical filter to overcome the interference from arc light and other light sources, and an embedded computer board for image processing and error calculation. By taking use of optical triangulation measurement method, the welding seam positions of cross direction and vertical direction can be measured from the feature points' positions of the welding seam in captured images. Advanced image processing algorithms are developed and implemented on the embedded computer board to process the images captured by the camera. Laser stripe region is defined from the entire image, and the effects of bright spots on the images caused by spatters and reflection are eliminated by area-size filtering algorithm. In order to detect the feature points, an algorithm based on the second order difference of the column indexes of the pixels on the laser stripe is implemented. Experiments results show that the measurement errors in cross direction and vertical direction are within the range of 0.5mm and 1.0mm respectively, which can meet the demands of accuracy for welding seam detection. The processing speed of the system is about 20 images per second, which can meet the demands of speed for arc welding robot.