Ensemble transform particle filter using regularized optimal transport and measure of nonlinearity. (November 2019)

• Record Type: Journal Article
• Authors: Kang, Chang Ho; Kim, Sun Young; Choe, Yeongkwon; Park, Chan Gook
• Is Part Of: Measurement. Volume 146(2019)
• Page Start: 363
• Page End: 371
• Publication Date: 2019-11
• Subjects: Ensemble transform particle filter -- Regularized optimal transport -- Degree of nonlinearity -- 2D LiDAR tracking; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2019.06.046
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Particle resampling based on linear programming is important in the ETPF. The resampling method is proposed with the ROTmap obtained by the Sinkhorn-Knopp algorithm. Measure of nonlinearity is to determine the need for the proposed resampling method. Improved efficiency of particle filter operation for application of 2D Lidar tracking system. Abstract: In this paper, the proposed method replaces a linear transformation step in an ensemble transform particle filter (ETPF) with an algorithm based on regularized optimal transport (ROT) resulting in improved performance. The linear transformation in the ETPF leads to warping of a posteriori weights into desired weights. However, it is computationally expensive and robust processing entails relaxation matching between posterior weights and desired weights. Therefore, a ROT method was used to address the ETPF challenges using a modified Sinkhorn-Knopp algorithm via the convergence of the process iteration rate. In addition, the proposed algorithm is correlated with the degree of nonlinearity such that the ROT method is only used for nonlinear measurement models to increase the efficiency of the proposed algorithm. In this paper, the proposed method is evaluated using 2D LiDAR via simulations and single object tracking experiments. Accordingly, the proposed method enhances the estimation of the target position compared to conventional methods.