Abstract: An approach is put forward to extract image feature points of fillet weld with the assistance of laser line. It overcomes the influence of the reflective stripes in the surface of fillet weld and identifies the feature points effectively. Firstly, the actual stripes and reflective stripes were distinguished by local contrast, then the actual stripes were divided by combination of morphological method and threshold segmentation, and the region of interest was determined. Secondly, the center points in light stripes were extracted according to the intensity distribution of the cross-section of light stripe. Finally, the iterative least-squares fitting method was applied to fit the piecewise centerline equations and to determine the fillet feature points. The experimental results show that this method can extract the sub-pixel image feature points from fillet weld with bright surface rapidly and accurately. It costs about 0.35s running on a PC whose CPU is clocked at 3.4 GHz, which can meet the real-time requirements of ordinary welding equipment with welding speed of 0-25 mm/s.