Comparative performance evaluation of GM-PHD filter in clutter

Random Finite Sets (RFS) offer a diligent formalism for tracking an unknown number of targets with multiple sensors. The Probability Hypothesis Density (PHD) filter, and its Gaussian Mixture (GM) and Sequential Monte Carlo (SMC) implementations, provide tractable Bayesian Filtering methods that propagate the first order moment of the RFS probability density. A feature of the PHD filters is that they do not require association to complete their correction step. This, we believe, should constitute a significant advantage, especially in scenarios of high false alarm rates and track intersections, which can easily compromise most observer-predictor methods that must perform association to carry out their correction step. To test this hypothesis, we compare the performance of the GM-PHD to the traditional Kalman (KF) and SMC filters for visual tracking of multiple targets in moderate to heavy false alarm rate scenarios. Our tracking and association performance results seem to support this hypothesis.

[1]  Ba-Ngu Vo,et al.  The GM-PHD Filter Multiple Target Tracker , 2006, 2006 9th International Conference on Information Fusion.

[2]  Sumeetpal S. Singh,et al.  Sequential monte carlo implementation of the phd filter for multi-target tracking , 2003, Sixth International Conference of Information Fusion, 2003. Proceedings of the.

[3]  R. Mahler,et al.  PHD filters of higher order in target number , 2006, IEEE Transactions on Aerospace and Electronic Systems.

[4]  Michael Isard,et al.  CONDENSATION—Conditional Density Propagation for Visual Tracking , 1998, International Journal of Computer Vision.

[5]  T. Başar,et al.  A New Approach to Linear Filtering and Prediction Problems , 2001 .

[6]  Nando de Freitas,et al.  An Introduction to Sequential Monte Carlo Methods , 2001, Sequential Monte Carlo Methods in Practice.

[7]  B. Vo,et al.  A closed-form solution for the probability hypothesis density filter , 2005, 2005 7th International Conference on Information Fusion.

[8]  D.E. Clark,et al.  An Efficient Track Management Scheme for the Gaussian-Mixture Probability Hypothesis Density Tracker , 2006, 2006 Fourth International Conference on Intelligent Sensing and Information Processing.

[9]  Jean-Marc Odobez,et al.  Evaluating Multi-Object Tracking , 2005, 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) - Workshops.

[10]  Ba-Ngu Vo,et al.  Probability hypothesis density filter versus multiple hypothesis tracking , 2004, SPIE Defense + Commercial Sensing.

[11]  R.P.S. Mahler,et al.  "Statistics 101" for multisensor, multitarget data fusion , 2004, IEEE Aerospace and Electronic Systems Magazine.

[12]  Ba-Ngu Vo,et al.  A Consistent Metric for Performance Evaluation of Multi-Object Filters , 2008, IEEE Transactions on Signal Processing.

[13]  R. Mahler.  Multitarget Bayes filtering via first-order multitarget moments , 2003 .

[14]  Ba-Ngu Vo,et al.  The Gaussian Mixture Probability Hypothesis Density Filter , 2006, IEEE Transactions on Signal Processing.