A real-time seam tracking system consisting of a three-axis cartesian robot, a line laser sensor, and an industrial computer was designed and its measuring principle, feature point measuring method and adaptive fuzzy control were researched. The Guassian Kernelized Correlation Filter(KCF) was adopted to detect the weld feature points in real time in welding process and 3D coordinate values of industrial camera were obtained by 2D pixel coordinate values based on measuring principle. A adaptive fuzzy controller was designed. On the basis of the adaptive fuzzy controller, the deviation values and deviation rate of the coordinate system were calculated to obtain the control values of movement trajectory for torch end. Meanwhile, the inputs and outputs on the domain and fuzzy membership functions of fuzzy controller were updated in real time. The experiments on seam tracking were performed. The experimental results show that the system realizes accurate seam tracking in real-time under the conditions of strong arc light and splash when the largest welding current is 350 A in Metal-Inert Gas Welding (MIG). The tracking error is less than 0.325 3 mm and the metrical frequency of sensor is up to 20 Hz. In welding processing, the end of the torch runs smooth the weld trajectory can be tracked accurately and the system shows strong anti-interference ability, so it meets the requirements of welding applications.