When solving the attitude of a driverless micro-electro-mechanical systems (MEMS) gyroscope, it is difficult to accurately assess the statistical characteristics of the system noise and the measurement noise due to the influence of the noise in the operating environment, which leads to a poor filtering effect. To this end, an adaptive optimization algorithm of Kalman filter parameters that can be applied in complex system environments is designed. Based on the demand for real-time and curve smoothness of the filtering algorithm for gyroscope attitude solving, the filtering real-time and curve curvature are used as evaluation functions, and an approximation iteration method is used to obtain the optimal system noise and measurement noise. The superiority of the algorithm is proved by mathematical simulation and semi-physical simulation test. The algorithm realizes the independent acquisition of the key parameters of the Kalman filter and improves the real-time filtering and the smoothness of the filter curve.