For the characterization of nonlinear and non-Gaussian, an estimation method fusing particle filter and predictive
filter is presented for the satellite attitude determination by using solely star sensor observations. At first, the system model
error parameter and attitude angular rate are estimated online based on predictive filter. Then the attitude quaternion is
estimated by the improved regularized particle filter. By designing proper particle initialization, important function, and
so on, the algorithm convergence rate is speeded up. The dimension of the particle filter is effectively reduced by fusing
predictive filter. A simulation is carried out on a general small satellite platform. Compared with the extended Kalman
filter(EKF), the result shows that the algorithm can converge fast and has good stable accuracy with respect to different initial
attitude conditions. The algorithm provides reference to particle filter design and attitude determination without gyro as well.