In order to meet the needs of flight test trajectory measurement, a highperformance flight test target trajectory measurement system is constructed. On the basis of this, an improved adaptive Kalman filtering method is proposed based on the multimachine learning to deal with the abnormal data in actual flight measurement. This method is to take the traditional unscented Kalman filter (UKF) as a basis. Firstly, BP neural network improved by genetic algorithm (GABPNN) is introduced to improve the UKF algorithm, realizing the regulation and correction of UKF global error, and improving the estimation accuracy of UKF. Furthermore, the outlier resistant technology is used to eliminate the isolated and spotted outliers in measurement, realizing the GABPNN The purpose of UKF's further improvement, and improving the robustness of filtering. Finally, the simulation used to verify the effectiveness of the new algorithm and the experimental analysis of the actual flight measurement data (the actual data obtained through the established trajectory measurement system) shows that the algorithm is valid.