Based on the spectral characteristics of GMAW arc,a monocular passive vision image acquisition system for molten pool was designed to capture the images of molten pool by adding a compound filter system and an external trigger system to the industrial camera. The average gray value of the image is used to judge the interference degree of arc light to the weld pool image,and the threshold value of wavelet transform modulus maximum edge detection algorithm is defined by the average gray value to extract the weld pool contour and its characteristic parameters. The molten pool contour features can reflect I-groove center and welding defects such as welding deviation（the deviation between groove center and wire projection center）,misalignment,excessive penetration and burn-through. The results show that the algorithm of molten contour extraction is reasonable,and the characteristic of molten pool defect is obvious,which provides technical support for the online monitoring of weld defects based on vision.