In order to better promote the continuous development of the industrial manufacturing, welding technology needs to be continuously innovated to improve efficiency and quality. The automatic welding seam tracking system based on structured light vision sensor can detect the welding seam image and extract the welding seam feature points in real time, thereby controlling the welding path and achieving intelligent welding. Therefore, the welding seam tracking system can better meet the needs of modern welding. However, the noise in the welding process seriously affects the process of weld image processing, which is not conducive to structured light stripes detection and weld feature points extraction. This thesis focuses on the line structured light image processing technology in the welding seam tracking process, so as to accurately detect the structured light stripes and weld feature points in the weld image under a complex noise environment. The main contents are as follows. Firstly, in order to accurately extract structured light stripes from weld images that contain noise such as smoke, spatter lines and arcs, a deep learning model combining semantic segmentation and object detection was proposed to detect the weld images. In the semantic segmentation branch, the model was optimized by adding parallel down-sampling modules and reducing the number of convolution kernels to increase the detection speed, and the feature extraction parts of this branch and the object detection branch shared the weights. Aiming at the problem that the proportion unbalance of structured light stripes and background pixels in the weld images caused the model segmentation results to be biased towards negative samples, we introduced a Dice coefficient into the loss function to correct the model. The experimental results show that the proposed method can achieve the extraction of structured light stripes with high accuracy on the basis of ensuring real-time performance. Secondly, on the basis of obtaining the structured light stripes, the center line of the laser stripes was extracted based on the principle of the center line method, the centroid method and the thinning method. After comparing the extraction accuracy, robustness, and real-time performance of the algorithms, the centroid method was chosen to extract the center line of the laser stripes. Then, the hough transform method was used to detect the straight line equation of the center line and find the intersection point, which was regarded as the welding seam feature point. Finally, using programming languages such as PyQt/C++/Opencv and deep learning model framework to design weld image processing software, integrate the above algorithms, and test them in actual welding scenarios. The experimental results show that the proposed welding seam image processing algorithm has better real-time performance. In addition, the proposed algorithm can obtain excellent detection results and accurately extract the welding seam feature points in a complex noise environment, so it has practical application significance.