By controlling the amount of feedback to the corresponding stable platform servo control mechanism to achieve stability control for stabilizing the platform, gyro is used to measure angular velocity of stabilized platform with respect to inertial space and then get the gesture of stabilized platform in inertial space. By analyzing the influence of gyro random error on the stabilized platform accuracy in detail, a processing method for gyro random drift was proposed. Firstly, owing to the values of the parameter α is artificially unreasonable of the Singer model in the field of motor tracking, the parameter of α in real time was estimated adaptively, and a new method of gyro modeling output module directly was established based Modified Singer (MS); secondly, on the basis of the established output model, an improved Kalman filter (IKF) based on interacting multiple model (IMM) was applied to the gyros output random error processing, and the theoretical analysis and derivation of the algorithm were explained in detail. Through numerical simulations and verification experiments of a certain type of gyro stabilized platform analysis, the static filter root mean square error is only 0.0227°/s, and this method was proved to be feasible and effective with the gyro drift treatments；finally, through the dynamic and static steady experiments of gyro stabilized platform, the MS-IMMIKF filtering algorithm was proved to be effective and practical for improving the accuracy of the stable platform. The dynamic and static steady experiments of gyro stabilized platform results show that the accuracy of the stable platform could be controlled within error of 3° and 25° respectively after the MS-IMMIKF algorithm filtering.