Feature Extraction of Welding Seam Image Based on Laser Vision

This paper addresses the feature extraction algorithm of the weld groove image for a welding robot system. The purpose is to improve the speed and the precision of the weld seam recognition. The acquisition of the weld image is a complex process, and consequently, it will be disturbed by a lot of noise. Specific methods must be used to process images. In this paper, the seam images with laser stripe under different types of groove are captured. After an analysis of the common methods in laser stripe center extraction, ridge line tracking, and direction template, which have not only excellent performance in the aspect of consuming time but also considerable line detection accuracy that agrees with the industrial application requirement well, are applied in the center extraction of stripe-shaped images. On the basis of the central line, the least square method based on the slope analysis method is used to detect the features of the image to obtain the feature information of the weld groove. Results of the validation are presented to demonstrate the accuracy of feature extraction. In addition, compared with the running time 212.708 ms of the direction template, the operation time of this paper is improved to 22 ms, which can meet the requirement of real time.