In the process of trajectory tracking measurement, the abnormal values of measurement data are in existence inevitably due to the complexity of the environment and the limitation of the measurement mechanism itself. Being ignorance from the influence of data quality on filtering accuracy, the traditional weighted observation fusion estimation algorithms are often directly used to deal with the measurement data from various sensors. For this reason, the robust estimation theory is introduced into the weighted observation fusion algorithm, and the measurement fusion residual vector, the robust weight factor and the equivalent covariance matrix of the fusion observation vector are calculated based on the observation fusion value and the fusion prediction value. The algorithm proposed realizes the realtime separation and correction of abnormal values, and solves the accuracy decreasing problem of the ballistic data fusion estimation due to the pollution of the measurement data. At the same time, the square root filtering idea is introduced, avoiding the filtering divergence problem caused by the nonpositive value of the error covariance matrix in the conventional UKF. The simulation results show that the fusion algorithm is high in estimation accuracy and has a little burden to computation, and can effectively reduce the influence of measurement errors on the accuracy of ballistic orbit determination.