本文
No.8(2013)1.Development of an outlier reduction filter
Yasuaki Kaneda, Yasuharu Irizuki
Recently, remote sensing methods, e.g., GPS, ultrasonic wave sensors, image measurements, radar measurements, and so on, are the focus of attention. However, noise from external environments often comprises part of the sensor signals and causes the normal Gaussian distribution to be distorted. In order to reduce effect of the outliers, some heuristic methods have been proposed, but design validation is difficult in these methods. In this research, we develop a method to reduce the outlier without heuristic design methods. The proposed method can be derived from an assumption that outliers are sparse and can be estimated by solving an optimization problem with ∫1 regularization. In addition, regularization parameters of the proposed method can be designed automatically by statistics of Gaussian measurement noise. The effectiveness is also demonstrated by some numerical simulations and experiments.
Keywords
Outlier, Robust Kalman filter, ∫1 regularization, Parameter design