Abstract: According to the special requirements of underwater automatic welding under environment of vision, a robotic underwater visual weld seam tracking system is developed.As underwater weld seam images are serious interference and fuzzy, wavelet transform technology is used to remove arc、splashing interference noise. An improved fuzzy enhancement image processing algorithm is adopted to improve the efficiency of image processing and obtain clear edge image.The weld seam center is picked up accurately by scanning the weld seam's left and right feature points after edge detection.In order to reduce tracking error, the coordinates offset of each small segment in weld track is divided into fifty parts, at the same time, an effective robot tracking control program is presented.The tracking experiment of sidelong, S-shape weld seam line showed that it can satisfy tracking demand under normal conditions.