Nowadays, three-dimensional measurement of highly reflective surface is an important part of industrial measurement. With the continuous development of automobile assembly, precision polishing, free-form surface processing and other industries, the demand for measurement of large size, large curvature and high reflective industrial parts such as aircraft tail, LCD panel, automobile body shell and windshield is increasingly strong. However, due to the large difference between the optical characteristics of the surface and the diffuse reflection object, it is difficult to achieve the ideal effect by the traditional three-dimensional measurement method.The phase deflection method is a non-contact high reflection surface 3D measurement method developed in recent years. Its high sensitivity and easy correction of system errors make it popular with many researchers. Generally, there are two main methods for the above measurement using deflection method: one is to improve the size of the screen and the field of view of the camera, but it is limited by accuracy and size. The other common method is to install a single-view system on the manipulator, and reconstruct a large size surface during the movement. However, the introduced manipulator system error reduces the accuracy, and the system mostly outputs the relative point cloud coordinates based on the gradient data, so the point cloud needs to be continuously spliced through the point cloud registration algorithm during the scanning process, so the calculation cost is high; however, this paper sets the camera array, and at the same time carries out the reconstruction based on absolute coordinates from multiple angles. It can effectively ensure the accuracy of local measurement, and realize accurate measurement of large size, large curvature and high reflection surface faster through the direct splicing of visual angles.In this paper, the LCD screen is used as the reference plane, and the camera array is used to realize the accurate reconstruction of the highly reflective surface shape based on multi-viewpoints. First, the mirror image of the reference plane target in the standard plane mirror is obtained using each camera, the external parameters of the reference plane and each camera through mirror calibration are obtained, and then the camera array is globally calibrated using the uniqueness of the reference plane. Then, the phase-shifting fringe are projected using the display screen to the measured surface, and the reflected modulated images are collected by each camera to obtain the absolute point cloud coordinates of the surface to be measured relative to each viewpoint. Finally, through global stitching, a larger size and higher curvature highly reflective surface shape measurement can be achieved compared to a single viewpoint system. At the same time, based on the planarity of the reference plane and the standard plane mirror, this paper proposes a coplanar constraint to optimize the geometric parameters and global parameters of the viewpoint and reduce the system error. The validity and accuracy of this method are verified by measuring the standard mirrors with different shapes.