Abstract: To solve the random drift error problem for MEMS inertial sensor in self-balancing two-wheeled electric vehicle gesture measuring, the EKF practical filtering algorithm is used to realize the information fusion of accelerometer and gyroscope. According to the experiment result of inertial sensor error characteristics, using Levenberg-Marquardt nonlinear least-squares iteration fitting data, and establishing the mathematical error models to compensate the random error of the acceleration sensor and the temperature drift error of the gyroscope, then the posture signal of self-balancing electric vehicle can be estimated optimally. The experimental based on REKF information fusion result shows that the self-balancing electric vehicle posture signal optimal estimation is effective and feasible, and it is beneficial to the two-wheeled electric vehicle self-balancing control.