The existing technology of the measurement of normal vector cannot meet the requirements of accuracy and timeliness. The methods of calculating normal vector cannot be applied to surfaces with different curvatures. In this paper, a method of normal vector measurement based on adaptive curvature is proposed by using binocular stereo vision. Firstly, based on the binocular stereo vision, the surface model of the hole point is established in accordance with the features of the variable curvature surface, and the layout method of the projected point is proposed. Then, according to the 3D coordinates of the projected points, the method of calculating normal vector based on the adaptive recognition of curvature is proposed. Finally, the measurement results of the small curvature surface part are compared with the normal vectors measured by CMM, to verify the accuracy of the method in this paper. The experimental results show that the error of the method is 1.6 °. The method can effectively improve the measurement accuracy of normal vector and satisfy the requirements of the in-site measurement of large aerial parts.