Real-Time Analysis of Labeled Multi-Bernoulli Trackers with Heavy-Tailed Measurements

This work considers the Joint Generalized Labeled Multi-Bernoulli filter and Labeled Poisson Multi-Bernoulli Mixture filter for heavy-tailed measurement noise distributions, e.g., the Student’s t distribution. These filters were used to track a team of micro aerial vehicles using a stereo vision camera in real-time. The measurement noise for the stereo vision camera were obtained by a sensor characterization procedure for a Gaussian and Student’s t distribution. This resulted in Gaussian mixture and Student’s t mixture versions of both filters. The performance of all four filters was assessed in terms of the Optimal Sub-Pattern Assignment(2) metric and computational time.