A three-dimensional visual positioning system based on line laser for fillet welds was independently developed.The 3D mathematical model and polynomial model of vision sensor were established by introducing line structured light plane.The conversion of pixel coordinates and world coordinates were achieved by calibrating the camera and the light plane.Step-by-step positioning was employed in the image processing section.Firstly,the feature area was coarsely positioned by means of template matching.Then,threshold image segmentation and morphological processing were used to obtain clear laser fringe images to reduce noise interference.Finally,fine positioning was carried out by bone extraction and straight line fitting to obtain accurate two-dimensional coordinates of feature points.Furthermore,combined with a three-dimensional model,3D world coordinates and positioning were achieved by coordinate transformation.One hundred images from different positions in the welding experiment were randomly selected for recording and comparing.Experimental results show that this method could effectively and accurately located the fillet weld.The average error is less than 0.15mm which can meet the accuracy requirements of welding owing to high precision.