Aiming at the requirement of high precision and fast fitting of surface features of parts in high-fidelity virtual assembly of head-up display (HUD) devices of automobiles and aircraft, a blue-ray random point cloud processing and fitting method is proposed. Firstly, the K-nearest neighbor point cloud combination filtering method is used to remove noise points and redundant points. Secondly, the internal surface of the filtered point cloud is quickly fitted based on the fast base surface construction and iterative optimization fitting method of non-uniform rational B-splines (NURBS). Finally, the point cloud boundary extraction and surface cutting are carried out based on the rolling ball algorithm to obtain high-precision surface features which could accurately reflect the point cloud boundary information. A complete set of HUD system standard parts was designed and processed, and the blue-ray point cloud of the main mirror was obtained to carry out fitting accuracy and efficiency evaluation experiments. By comparing the fitting surfaces of the proposed method, commercial software method and literature method with the surface data obtained by three coordinates accordingly, it can be concluded that the proposed method can achieve micron fitting accuracy on the premise of significantly improved fitting efficiency, and can meet the technical requirements of high-fidelity virtual assembly.