Study on Processing for Seam Image Based on CCD Vision Sensing

In automatically seam tracking system with structured light vision sensor in girth-welding of pipeline, there is a lot of noise in seam images. To extract seam information reliably, it is important to process the original images to eliminate influence of noise. A novel compounding process method is presented. Firstly, one dimension Laplacian of Gaussian filter was applied to transform noises whose width was larger than laser light stripe to isolated and small noise. Then the image was smoothed with neighborhood median filter. Due to gray distribution variety of seam image, binary image was created with maximal square difference adaptive threshold within selected section based on priori information. Acnode noise was cleared up by region labeling and selected threshold, consequently a binary image of seam is obtained. Finally, smooth and consecutive skeleton of seam is extracted with morphologic thinning method. Experiments indicate that this process method satisfies requirement of real-time seam tracking. Key-words:Vision sensor, Seam tracking, Filter, Image processing, CCD, Structured light