Vehicle head-up system is the inevitable trend of the development of vehicle instrument display in the future. However，in the process of free-form mirror measurement in head-up display system， the features of different regions are similar， which leads to be difficult to extract feature points， and to fall into local minimum value during optimization. A high-precision method for matching the three-dimensional point cloud with the head-up display mirror surface model is proposed in this paper. Firstly， the measurement data and model are converted into point cloud files， and after three-dimensional rotation， the normal vector of each point cloud is parallel to the Z axis， and then point cloud is projected to the XOY plane， which will not cause overlap and occlusion. Secondly， edge points are extracted in the two-dimensional plane， and the corner points of the edge are extracted. The rotation angle is calculated by determining the corresponding corner points， and then the corresponding rotation of point clouds is completed in the three-dimensional space. Thirdly， the rough matching is completed， and the fine matching is completed by minimizing the objective function. The algorithm in this paper uses the edge features of free-form surface mirrors for matching， avoids extracting local feature points， and does not fall into local minimums during the optimization process， and can accurately calculate the error of the tested surface.