Abstract: An exact rover localization algorithm in bundle adjustment (BA) was proposed， which used LevenbergMarquardt algorithm instead of GaussNewton algorithm to obtain highprecision localization information. It could be used especially in the cases of lacking adequate control information in the lunar surface environment. According to the core ideas of the LevenbergMarquardt algorithm and lunar image structure network， the error estimation based on bundle adjustment was formulated， and then the appropriate damping strategy and posterior weights estimation were given. Simulation results showed that the disadvantages of weak applicability of GaussNewton algorithm could be overcome by using the proposed algorithm， and it has a high localization accuracy and iterative convergence rate.

Key words: bundle adjustment, GaussNewton algorithm, LevenbergMarquardt algorithm, navigation and localization, posterior weights estimation