The Outlier Elimination of Trajectory Data Based on Kernel Adaptive Filtering with Variable Step Size

An outlier detection and elimination method based on kernel adaptive filtering with variable step size for trajectory data of vehicle test was proposed. The training sample of kernel adaptive filter is designed according to the effective trajectory data. After training, the residual error between the output of the kernel filter and the trajectory can be obtained. If the residual error at some time point is larger than 3 times of the standard deviation of the residual error, the corresponding data point can be considered to be the outlier data based on Wright guidelines, and then the data is instead of the output of the kernel adaptive filter to eliminate the outlier data. To further improve the precision of the outlier data elimination and interpolation, a variable step size algorithm was designed according to the output error of the kernel adaptive filter, in which the step size can be controlled during the iterative process. The proposed method can implement outlier data elimination and interpolation at the same time, which has good robustness and high precision. The simulation and test data processing results show the effectiveness.