A novel hypothesis splitting method implementation for multi-hypothesis filters

Abstract

The paper presents a multi-hypothesis filter library featuring a novel method for splitting Gaussians into ones with smaller variances. The library is written in C++ for high performance and the source code is open and free1. The multi-hypothesis filters commonly approximate the distribution transformations better, if the covariances of the individual hypotheses are sufficiently small. We propose a look-up table based method to calculate a set of Gaussian hypotheses approximating a wider Gaussian in order to improve the filter approximation. Python bindings for the library are also provided for fast prototyping.