Fast recognition algorithm of weld seam image under passive vision

For common butt welds and poor image quality of weld collection, this paper designed a fast recognition algorithm of weld based on passive vision in the case of strong arc. Multiple recognition threads are managed by calling one image recognition management thread to improve the rate of image recognition. The recognition window is selected to reduce the complexity of the image, and according to the recognition results, the initial position of the recognition window for the next frame is fed back. Through the adaptive threshold segmentation method, the image is threshold segmentation, and then the connected domain analysis is carried out to judge the target image. Compared with the traditional single threshold segmentation method, this method can segment the target image area more accurately. According to the identified target image area refinement recognition window and the iterative recognition process, the more accurate target image is segmented, and then the weld skeleton is refined by the refinement algorithm.