Abstract: A new approach to measure the deviation of seam tracking was proposed based on near-infrared image processing and recognition during high-power fiber laser welding.The molten pool images were acquired by a high speed camera with the narrow bandpass filters.Based on the gray distribution of the near-infrared image of the molten pool,its temperature distribution parameter was defined as the measured eigenvalue of seam tracking.A near-infrared image model was established to measure the deviation between the welding seam and the laser beam focus in real-time by using the least-square algorithm.The effectiveness of the proposed approach was verified by the welding experiments.