Position Error Modeling Using Gaussian Mixture Distributions With Application to Comparison of Tracking Algorithms

In this paper Gaussian mixtures are used to model the distribution of position error in tracking algorithms. An expectation maximization algorithm is constructed to estimate parameters of a k-component Gaussian mixture based on a sample set obtained from a tracking simulator. The modeling and parameter estimation approach is applied to position error data generated by several tracking algorithms including multi-target multi-sensor joint probabilistic data association and particle filters. The Gaussian mixture model yields significantly better likelihood of position error compared to the single Gaussian distribution. It is further shown how the mixture models can be used to efficiently compare tracking algorithms in terms of the root mean squared position error.