Abstract: Objective The weld width is one of the important indicators to characterizing welding quality and welding stability. This paper aims to improve the problem of high difficulty and error in measuring weld width of fillet weld. Methods The paper designs an image acquisition method for fillet weld, proposes a method based on passive vision to obtain weld width, and improves the edge extraction algorithm of traditional Canny operator. The images of weld are captured by using CCD industrial camera, the captured images are preprocessed by ROI extraction, grayscale, median filtering, bilateral filtering and Gaussian filtering, the edge of weld width is extracted by improved Canny operator, the melt pool contour line segment is fitted by probabilistic Hough transform, and the weld width is obtained by using normalized Euclidean equation. Results The experimental results show that average measurement time of weld width is 160 ms, and average absolute error is 0.34 mm. Conclusion The image processing speed and measurement accuracy of this method meet the requirements of weld surface detection, and the robustness is strong.