A new improved filter for target tracking: compressed iterative particle filter

Hongbo Zhu, Hai Zhao, Dan Liu, Chunhe Song

.

DOI: 10.4236/ns.2011.34039                

Abstract

Target tracking in video is a hot topic in computer vision field, which has wide applications in surveillance, robot navigation and human-machine interaction etc. Meanshift is widely used algorithm in video target tracking field. The basic mean shift algorithm only considers the color of targets as the tracking characteris- tic feature, so if the appearance of the target changes greatly or there exits other objects whose color is similar to the target, the tracking process will fail. To enhance the stability and robustness of the algorithm, we introduce par- ticle filter into the tracking process. Basic particle filter has some disadvantages such as low accuracy, high computational complexity. In this paper, an improved particle filter GA-UPF was proposed, in which a new re-sampling algorithm was used to predict target centroid position. The target tracking system of binocular stereo vision is designed and implemented. Experi- mental results have shown that our algorithm can tracking object in video with high accuracy and low computational complexity.

Keywords

Filtering Method; Particle Filtering; Unscented Particle Filter; Genetic Algorithm; Target Tracking

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]	Julier S J, Uhlmann J K. A general method for approxi- mating nonlinear transformation of probability distribu- tions [EB/OL], http://www.robots.ox.ac.uk/~siju/work/ publications/Unscented.zip, 1997-09-27
[2]	ZhiQiangHou, ChongZhaoHan. Based on the background pixel classification algorithm of reconstructing [J], Jour- nal of software, 2005, 16(9): 1568-1576.
[3]	Pitt M K, Shephard N. Filtering via simulation: Auxiliary particle filters [J]. Journal of the American Statistical Association, 1999, 94(2): 590-599.
[4]	Berzuini C, Best N. Dynamic conditional independence models and Markov chain Monte Carlomethods[J], J of the American Statistical Association, 1997, 92(5): 1403- 1412.
[5]	Belviken E, Acklam P J. Monte Carlo filters for non-linear state estimation [J], Automatica, 2001, 37(1): 177-183.
[6]	Doucet A, Gordon N. Sequential Monte Carlo methods in practice [M], New York: Springer-Verlag, 2001.
[7]	Lei Lei, Yan-jun Li, Ke Zhang. A Fast Algorithm for Searching Object Centriods in Binary Images [J], Infra- red Technology, 2007, 29(9): 548-551.
[8]	Higuchi T. Monte Carlo filtering using genetics algo- rithm operator [J], J of Statistical Computation and Simulation, 1997, 59(1): 1-23.
[9]	Mo Y W, Xiao D Y. Hybrid system monitoring and diag- nosing based on particle filter algorithm [J], Acta Auto- mation Sinica, 2003, 29(3): 641-648.
[10]	Freitas J F G, Niranjan M, Gee A H,et al. Sequential Monte Carlo Methods to Train Neural Network Mod- els[J], Neural Computation, 2000, 12(4): 995-993.
[11]	der Merwe R V, Doucet A. The Unscented Particle Filter [DB/OL], http://cslu. cse. ogi. edu/publications/ps/UPF_ CSLU_talk.pdf
[12]	LI L Q, JI H B, LUO J H, Iterated extended kalman par- ticle filtering[J], Journal of XIDIAN university (Natural Science), 2007, 34(2), 1172-1175.
[13]	ZHANG Shuchun, HU Guangda, Variations of unscented Kalman filter with their applications in target tracking on re-enty [C], Proc of IEEE, the 25th Chin Control Conf, 2006: 407-412.
[14]	GUO W Y, HAN C Z, LEI M, Particle distribution con- trol for an iterated unscented kalman particle filter [J], Journal of Tsinghua University (Sci&Tech), 2007, 47(2): 1866-1869.
[15]	WANG F S, ZHAO Q J, A new particle filter for nonlin- ear filtering problems [J], Chinese Journal of Computers, 2008, 31(2).
[16]	Calballero-Gil P, Fuster-Sabater A. A wide family of non- linear filter function with a large linear span [J], Informa- tion Science, 2003, 164(1-4): 197-207.