A modified Square-root Unscented Kalman Filter restraining outliers

Aiming at solving the problem that the accuracy and stability of Square-root Unscented Kalman Filter(SRUKF) will be affected if there are outliers in the observation values, this paper proposes an improved SRUKF restraining outliers based on orthogonality of innovation in the filtering processing. This modified filter at first detects whether there are outliers in the observation values by judging if the orthogonality of innovation is lost or not, and then assigns an activation function as the weight to each observation value, which can keep the orthogonal properties of the innovation sequence and the outliers can be detected and corrected. The Matlab simulation results about the GPS/DR integrated navigation system show that this modified filter is effectively resistant to outliers in the observation values and improves the filtering accuracy and stability.