Thin plates constitute more than 80% of the structural components in shipbuilding, and butt welding is an essential technological process in ship manufacturing. Different from general butt welding, thin plate welding for shipbuilding is characterized by long length (≥ 15 m), narrow seam width (≤ 1 mm), and high precision demands (≤ 0.5 mm). The clamping deformation, arc sparks, and noise pollution during the welding seriously affect the tracking precision in automated welding. To address this problem, this paper presents a seam tracking algorithm combining image processing techniques and spatiotemporal tracking data filtering techniques to acquire a steady and smooth torch trajectory with less position variation. In the image processing stage, a spatiotemporal density mask image processing method is proposed to preserve effective pixels in spatiotemporal laser images to resist noises and improve the accuracy of welding point position estimation. Experimental results show that the frequency of the proposed seam tracking method can reach 30 Hz, the average absolute tracking error is less than 0.25 mm, and the maximum tracking error is less than 0.4 mm. The welding torch runs more smoothly during welding thus leading to higher welding quality.