The invention discloses a complex fillet weld joint position detecting method based on laser binocular vision. The complex fillet weld joint position detecting method based on the laser binocular vision overcomes the defect that a traditional image detecting algorithm is poor in extracting precision and extracting stability of fillet weld joint feature points. According to the technical scheme, the complex fillet weld joint position detecting method based on the laser binocular vision includes the following steps that a laser scanning workpiece is used, a binocular CCD camera is used for collecting an image, and the collected grey level image is transmitted to a software system of a computer to be processed; the processing procedures are sequentially median filtering, binarization, refining, obtaining the maximum connected domain, close angular point searching, straight line fitting and intersection point obtaining; finally, image coordinates of the weld joint feature points are obtained. The pre-processing method (median filtering, binarization, refining, obtaining the maximum connected domain) is used for removing interference of the background of the image and narrowing the retrieval range of sequential processing. A close angular point searching strategy is provided, the speed of angular point detection is improved and finally, the detection precision reaches the sub-pixel class by means of the straight line fitting and intersection point obtaining method.