Research Of Attitude Estimation Algorithm Based On Quaternion Fast Particle Filter

With the improvement of the computing power of modern computers,the method of using Monte Carlo filtering methods to solve the nonlinear problems in attitude estimation has gradually matured,but the attitude filtering algorithms that are more suitable for modern computer multi-core parallel processing have not been studied accordingly.In this paper,a micro-electromechanical inertial sensor is used as the basic component of the attitude measurement device of the aircraft.According to the results of Allan variance analysis,an improved algorithm using new quaternion distribution is presented to cope with the huge computation burden of quaternion particle filter in aircraft attitude estimation.First,this paper analyzed and summarized the parallel processing development of attitude estimation algorithms,attitude deterministic algorithms,and particle filtering at home and abroad,and classifies them according to the corresponding theoretical basis of each algorithm which determines the direction of parallelization improvement of pose estimation algorithm in this paper.Secondly,this paper selected the corresponding reference coordinate system,and compares the advantages and disadvantages of various attitude description parameters.Quaternions were selected as the attitude description parameters of the aircraft.The Allan variance method was used to analyze the noise of the MEMS inertial sensor and identify the noise composition of the sensor and the corresponding noise standard deviation,according to which,sensor error model was established.Then verify the correctness of the established sensor error model by using this method.Thirdly,this paper did some research on Bayesian estimation methods.The application of particle filtering in attitude estimation is studied,and the superiority of quaternion particle filtering is discussed.The modified Gaussian particle filter is used to estimate the gyro biases that play an important role in aircraft attitude.For the initialize produce of quaternion particle filter,an estimation method that adaptively adjusts the particle set size according to the distribution of state estimation is proposed,so that the filtering is in the case of noise changes which avoid filtering estimation divergence.Finally,this article first introduced the quaternion random distribution to make parallelization improvements to quaternion particle filtering,approximated the state posterior distribution to a Gaussian distribution on the quaternion hypersphere,and used linear transformation to update the state to obtain prior distribution of the state,which accelerates the calculation speed.Corresponding simulation research is carried out on the proposed algorithm,and the results show that the proposed algorithm improves the average calculation speed by nearly 50% compared to quaternion particle filtering,and is more stable and faster than the initial convergence of EKF and UKF.In this paper,a flight platform and an embedded platform are built accordingly,and the feasibility of the proposed algorithm to implement attitude estimation in an embedded multi-core CPU is verified.