Stereovision-based Initial Pose Estimation Relative to Non-cooperative Space Target

Initial relative parameters are of great importance as inputs of the subsequent filter process based on the stereovision system. The relative dynamic model is approximated to estimate the initial relative state. Firstly, the error distribution property of feature point captured using cameras can be approximately regarded to be zero-mean Gaussian distribution when the measurement error of pixel is assumed to be the same. Furthermore, the cofactor matrix can be derived to describe the error distribution property. Built on what is mentioned above, the weighted total least square is adopted to compute the relative angular velocity. To analyse the estimation precision of the algorithm, the Cramer-Rao low bound is also exploited to calculate the theoretical precision trend with the number of the feature point. The relative position estimation uses the particle swarm optimisation algorithm to seek optimal solution due to the fact that the corresponding coefficient matrix is not full rank. Finally, three groups of simulation experiments are conducted to verify its effectiveness. The experiment results show that this algorithm can be utilised to estimate the initial relative state.