Research On Seam Image Processing Based On Laser Vision Sensing

Seam tracking is an important element to realize welding automation, while image processing is an important element to realize seam tracking. Apart from assembly and machining, welding is one of the important areas in modern manufacturing process. Conventional welding techniques are mostly done manually, how to improve the efficiency and quality of steel structure welding and reduce the labor intensity of welders have become a key issue to be solved. Hence, research and application of welding automation and intelligence has become one of the major topics in modern industrial production. Based on the domestic and international related research on the seam image processing this dissertation taking practical applications into consideration, the research target of this paper is to achieve automatic seam tracking in terms of laser vision, and the seam image processing methods in every steps are studied in detail. The main contributions of this paper are summarized as follows:(1) By combining adaptive median filtering and clustering algorithm and making some modifies, an adaptive clustering median filtering algorithm is proposed for removing high-density salt and pepper noise, which shows superior performance on removing noises. For random noise, further modified is made to process the high-density random noise effectively.(2) To process the filtered seam image, a segment clustering algorithm is proposed for segmenting the histogram of filtered seam image, which enhances image contrast and reserves regional details simultaneously. Otsu algorithm can be used to find the threshold of seam image binarization in the histogram segmentation of laser region, thereby realizing the adaptive threshold selection according to different seam image.(3) The algorithm of anti-noise-dilation-erosion morphological edge detection is be used to detect the edge of preprocessed seam image. Using a circular structure element with radius decided by depth of the laser with a binary image area, the edge of seam image can be detected while the spatter and noise are effectively elimination.(4) The local centerline is effectively extracted by quantifying the depth of the target area. Then, we expanding it to get the centerline of laser region from shallow to deep. This method can effectively thinning the various target image and the algorithm is simple and easy to implement. Subsequently, the SUSAN algorithm is introduced to effectively detect the feature points of groove and seam.