Abstract: Pointing measurement runs through each stage of satellite assembly, integration and test (AIT), and pointing accuracy directly affects the in-orbit service performance of satellite. With the development of high quality, high benefit and high efficiency in the space field and the large-scale and diversified manufacturing mode of commercial space, the satellite manufacturing cycle is getting shorter and shorter, and the manufacturing quality is getting higher and higher. The traditional multi-theodolite human industrial station manual coligning method has the shortcomings of low measurement efficiency, low measurement accuracy, high labor intensity and low degree of automation, which can not meet the development trend of aerospace field. It is urgent to develop an automatic pointing measurement system with high precision for on-board equipment. The difficulty of this system lies in the recognition and normal estimation of mirror objects in unstructured environment (such as low illumination, low contrast, etc.) during the rough collimation process. In this paper, a system and method for recognition and normal estimation of mirror objects based on reflection are presented. Based on the specular reflection characteristics of the target cube mirror, the normal information of the cube mirror is accurately estimated based on the specular reflection principle, the parameter estimation of the homography matrix, and the coordinates of the reflection points on the reflection screen. The effectiveness and accuracy of the proposed method are verified by experiments.