Abstract: To address the problems of low efficiency and poor accuracy of existing weld quality detection methods, a weld quality detection and evaluation method based on the binocular phase-shifting method was proposed. The point clouds of the weldment with a weld were first obtained based on the four-step phase-shifting method and the principle of binocular vision. Secondly, a model combining “rough extraction” and “fine extraction” was adopted to extract the weld point clouds from the weldment point clouds. Then, the ridge line of the weld point clouds was extracted; the direction line of the ridge line was fitted by the cubic B-spline curve fitting algorithm, and the cross-section point clouds of the weld were extracted accordingly. Finally, a multi-dimensional quantitative evaluation index system of weld quality was constructed, and the objective evaluation of weld forming quality was completed based on the geometric characteristics of the cross-section point clouds. A weld point cloud acquisition system was established, and experiments were conducted. The maximum deviation of the weld width obtained in the experiments is 0.287 mm, and the minimum deviation is 0.105 mm. The experimental results indicate that the proposed method can be applied to the detection and evaluation of various types of welds and has the advantages of strong stability and high precision, which can meet the application requirements of high-precision weld quality detection and evaluation in the industrial field.