Abstract: In order to obtain weld pool images with better quality information, a beam splitter was used to transmit the single light source to two cameras. With the synchronous triggering device, a synchronous acquisition system of visible and near infrared weld pool images was established. Due to the different deformations between the images of the two cameras, an image registration method is proposed based on affine transformation to registere the visible and near infrared images of the weld pool. Then the image fusion method is also proposed based on principal component analysis. Compared with the traditional image fusion methods, weighted average fusion, IHS fusion and wavelet fusion, the principal component analysis image fusion method is the most effective and the best weld pool image quality is obtained.