Loading...

Multitarget Tracking with Improved Particle Filter Eliminating Data Association Step

Raees Danaee, Meysam | 2013

845 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 44342 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Behnia, Fereidoon
  7. Abstract:
  8. In general, multi-target tracking consists of estimation of the posterior density function of present targets at each scan in the observation area. These targets may have unknown and time varying number of targets. It is a tough job due to misdetections, false alarms, data association ambiguity, and nonlinear equations-non Gaussian noises. These all make it difficult to apply Kalman filter and its extensions such as extended Kalman filter and unscented Kalman filter. Monte Carlo methods, particularly particle filters, have recently aroused the interest of designers and enjoyed a lot of success to deal with multi-target tracking difficulties. In addition, they can handle nonthresholded data as well as non ideal thresholded measurements such as unresolved data which is impossible for classic multi-target tracking systems. Classic tracking systems give way to Monte Carlo methods, yet there is a possibility to upgrade Monte Carlo methods further down the road. In this dissertation, as it is assigned in the PhD proposal, we pursue two goals. At first, we extend particle filter to cope with multi-target tracking in environments with varying number of targets. Second, for scenarios with the fixed and known number of targets, we try to improve particle filter performance so that tracking accuracy and convergence rate increase for the less number of targets. Furthermore, it is tried to suggest algorithms which have the capability to successfully deal with nonthresholded data or non ideal thresholded measurements. In each chapter, simulation results demonstrate the superiority of the proposed algorithms compared to the methods undertaken at the cutting edge of the related Monte Carlo fields
  9. Keywords:
  10. Particle Filter ; Multitarget Tracking ; Auxiliary Variable Particle Filter ; Track-Before-Detect ; Cardinalized Probability Hypothesis Density

 Digital Object List

 Bookmark

No TOC