文摘
Traditionally, automated quality inspection of welding tasks relies on nonvisual information and is mainly done off-line. In this work, we introduce an image acquisition system which is capable of monitoring the welding process on-line, resulting in high-quality image information during an ongoing welding process. We show how to further exploit this image information by automatically tracking the weld seam position in the image, even under heavy smoke and gas disturbances. We exploit the high information redundancy between subsequent frames given by large overlap to generate a seamless image of the entire weld seam and effectively suppress adverse optical effects caused by, e.g., smoke and sparks.