VINS (visual-inertial navigation systems) solve a series of parameters required for state estimation through initialization, such as scale, gravity vector, velocity, inertial measurement unit′s bias, etc., to improve the accuracy of navigation positioning and environmental perception of the system. The initialization methods of VINS can be divided into three categories according to the sensing information fusing mode:joint initialization, disjoint initialization and semi-joint initialization. Based on the existing research work, the current mainstream initialization methods of VINS were reviewed from four aspects:basic theory, development and classification, existing methods, performance evaluation, and the future development trends were summarized, which is helpful to have a general understanding of VINS initialization methods and grasp its development direction.