The relative navigation approach for UAVs formation was presented which considered vision sensors’ error statistics following non-Guassian probability distributions. The relative inertial equation and measurement principal of line of sight were provided. The Huber-based filtering was applied to fuse the relative inertial information and line of sight, as a result, the relative attitude, relative position and relative velocity among UAVs were estimated. The Huber-based filtering was a combined minimum and norm estimator which could exhibit robustness for the perturbed Guassion probability distribution. The effectiveness of relative navi-gation approach could be proved by simulation result.