The MEMS IMU has the advantages of small size, lightweight, low cost, and high reliability. It is widely used in many fields such as robotics, virtual reality and smart wear. Low-cost MEMS inertial measurement units are affected by noise and zero-bias errors in practical deployment, hence testing and error compensation methods are required to improve their actual use accuracy. This article proposes a method that comprehensively tests and compensates for the error in the inertial measurement unit. Firstly, the error model of MEMS-IMU is established, and the deterministic parameters in the error model are calibrated by the optimization method. Secondly, the Allan variance analysis method is utilized to calibrate the random error parameters. Finally, the nonlinear optimization method fused with vision is used to estimate and compensate the zero bias online and in real-time, thereby achieving the goal of improving the navigation and positioning accuracy of the MEMS-IMU. Through experimental analysis, the above combined method does not need to use the specific test and calibration equipment, and can effectively compensate for the error of the low-cost MEMS inertial measurement unit and improve the positioning accuracy.