In order to improve the centroid location accuracy of fine guidance sensor in dynamic environment, a two-step restoration method is proposed to solve the problem of star spot trailing caused by angular motion of carrier and small amplitude random vibration. Constrained least squares filtering is used to eliminate the long tail of the star spot caused by the angular motion of the carrier according to the fuzzy kernel function detected by the two fast Fourier transforms of the blurred star spot. Aiming at the residual blurring caused by small amplitude random vibration of the carrier, the clear star spot gradient distribution prior is used as a regular constraint to iteratively blind restoration of the coarse restoration results. The half-quadratic optimization algorithm is introduced to solve the non-convex cost function to improve the convergence speed of the iteration. The experimental results show that the restored star spot is close to the Gauss distribution in the dynamic environment of 4000 μrad/s angular velocity, and compared with the inverse filtering and R-L methods, the peak signal-to-noise ratio is 61.9% and 32.9% higher and the error of centroid location is 59.9% and 43.4% lower, respectively.