韓玉蘭 韓崇昭
摘 要:目前擴(kuò)展目標(biāo)跟蹤算法大都假設(shè)其系統(tǒng)為線性高斯系統(tǒng),針對(duì)非線性系統(tǒng)的多擴(kuò)展目標(biāo)跟蹤問(wèn)題,提出了采用粒子濾波技術(shù)對(duì)目標(biāo)狀態(tài)和關(guān)聯(lián)假設(shè)進(jìn)行聯(lián)合估計(jì)的多擴(kuò)展目標(biāo)跟蹤算法。首先,提出了將多擴(kuò)展目標(biāo)狀態(tài)和關(guān)聯(lián)假設(shè)進(jìn)行聯(lián)合估計(jì)的思想,解決了在估計(jì)目標(biāo)狀態(tài)和數(shù)據(jù)關(guān)聯(lián)時(shí)相互牽制的問(wèn)題;其次,根據(jù)擴(kuò)展目標(biāo)演化模型、量測(cè)模型建立多擴(kuò)展目標(biāo)狀態(tài)和關(guān)聯(lián)假設(shè)的聯(lián)合建議分布函數(shù),并利用粒子濾波技術(shù)實(shí)現(xiàn)聯(lián)合估計(jì)的Bayes框架;最后,為解決直接采用粒子濾波實(shí)現(xiàn)時(shí)存在的維數(shù)災(zāi)難問(wèn)題,將目標(biāo)聯(lián)合狀態(tài)粒子的產(chǎn)生和演化分解為各個(gè)目標(biāo)狀態(tài)粒子的產(chǎn)生和演化,對(duì)每個(gè)目標(biāo)的粒子集根據(jù)與其相關(guān)的權(quán)重單獨(dú)進(jìn)行重抽樣,這樣在抑制目標(biāo)狀態(tài)估計(jì)較差部分的同時(shí)使每個(gè)目標(biāo)都保留了對(duì)其狀態(tài)估計(jì)較好的粒子。仿真實(shí)驗(yàn)結(jié)果表明,與擴(kuò)展目標(biāo)概率假設(shè)密度濾波器的高斯混合實(shí)現(xiàn)方式和序貫蒙特卡洛實(shí)現(xiàn)方式相比,所提算法的狀態(tài)估計(jì)精度較高,形狀估計(jì)的Jaccard距離分別降低了30%、20%左右,更適合于非線性系統(tǒng)的多擴(kuò)展目標(biāo)跟蹤。
關(guān)鍵詞:擴(kuò)展目標(biāo)跟蹤;非線性系統(tǒng);Bayes框架;聯(lián)合估計(jì);粒子濾波;建議分布函數(shù)
中圖分類號(hào):TN273
文獻(xiàn)標(biāo)志碼:A
Abstract: Most of current extended target tracking algorithms assume that its system is linear Gaussian system. To track multiple extended targets for nonlinear Gaussian system, an multiple extended target tracking algorithm using particle filter to jointly estimate target state and association hypothesis was proposed. Firstly, the idea of joint estimation of the multiple extended target state and association hypothesis was proposed, which avoided mutual constraints in estimating target state and data association. Then, based on extended target state evolution model and measurement model, a joint proposal distribution function for multiple extended target and association hypothesis was established, and the Bayesian framework for the joint estimation was implemented by particle filtering. Finally, to avoid the dimension disaster problem in the implementation of the particle filter, the generation and evolution of the multiple extended target combined state particles were decomposed into that of the individual target state particles, and the particle set of each target was resampled according to the weight association with it, so that each target retained the particles with better state estimation while suppressing the poor part of target state estimation. Simulation results show that, in comparison with the Gaussianmixture implementation of extended target probability hypothesis density filter and the sequential Monte Carlo implementation of that, the estimation accuracy of the target state is improved, and the Jaccard distance of shape estimation is reduced by approximately 30% and 20% respectively. The proposed algorithm is more suitable for multiple extended target tracking of the nonlinear system.
英文關(guān)鍵詞Key words: extended target tracking; nonlinear system; Bayesian framework; joint estimation; particle filter; proposal distribution function
0 引言
擴(kuò)展目標(biāo)在每個(gè)時(shí)刻可產(chǎn)生多個(gè)量測(cè),因此傳統(tǒng)多點(diǎn)目標(biāo)跟蹤算法無(wú)法應(yīng)用于多擴(kuò)展目標(biāo)跟蹤。目前多擴(kuò)展目標(biāo)跟蹤算法大致有兩類: 一類是通過(guò)修改假設(shè)條件將點(diǎn)目標(biāo)跟蹤算法的數(shù)據(jù)關(guān)聯(lián)方法如聯(lián)合概率數(shù)據(jù)關(guān)聯(lián)(Joint Probabilistic Data Association, JPDA)、概率多假設(shè)方法(Probabilistic MultiHypothesis, PMHT)等,推廣到多擴(kuò)展目標(biāo)跟蹤[1-3];另一類是基于隨機(jī)有限集,將概率假設(shè)密度(Probability Hypothesis Density, PHD)濾波器、勢(shì)概率假設(shè)密度(Cardinalized PHD, CPHD)濾波器、高斯混合概率假設(shè)密度(Gaussian Mixture PHD, GMPHD)濾波器、序貫蒙特卡洛概率假設(shè)密度(Sequential Monte Carlo PHD, SMCPHD)濾波器等應(yīng)用到多擴(kuò)展目標(biāo)跟蹤算法[4-7],但這類算法理論上需要考慮每一時(shí)刻量測(cè)集的所有可能劃分,因此計(jì)算量較大,計(jì)算量會(huì)隨著擴(kuò)展目標(biāo)個(gè)數(shù)或量測(cè)個(gè)數(shù)急劇增加。文獻(xiàn)[6-8]為減少計(jì)算量只考慮了一部分劃分算法,但擴(kuò)展目標(biāo)跟蹤性能嚴(yán)重依賴于劃分算法,在目標(biāo)相距較近時(shí)難以獲得理想的效果。
目前已存在非線性系統(tǒng)的單擴(kuò)展目標(biāo)跟蹤算法,如文獻(xiàn)[9]中將RaoBlackwellised粒子濾波器應(yīng)用到擴(kuò)展目標(biāo)跟蹤,線性狀態(tài)部分采用卡爾曼濾波器,非線性部分采用粒子濾波器進(jìn)行估計(jì);文獻(xiàn)[10]中將非線性量測(cè)函數(shù)線性化,利用基于隨機(jī)矩陣的擴(kuò)展目標(biāo)跟蹤算法擴(kuò)展到非線性系統(tǒng)?,F(xiàn)有的多擴(kuò)展目標(biāo)跟蹤算法一般是針對(duì)線性高斯系統(tǒng),為解決非線性問(wèn)題通常將處理非線性系統(tǒng)的方法如無(wú)跡卡爾曼濾波器(Unscented Kalman Filter, UKF)、粒子濾波器(Particle Filter, PF)與線性系統(tǒng)的多擴(kuò)展目標(biāo)濾波器相結(jié)合,如文獻(xiàn)[11]中將UKF應(yīng)用于擴(kuò)展目標(biāo)GMPHD(Extended Target GMPHD, ETGMPHD)濾波器,采用非線性量測(cè)模型實(shí)現(xiàn)狀態(tài)估計(jì)的更新,但是這種處理非線性的方式的濾波性能會(huì)隨著非線性程度的增加急速下降。
本文針對(duì)多擴(kuò)展目標(biāo)跟蹤的數(shù)據(jù)關(guān)聯(lián)和非線性問(wèn)題,由擴(kuò)展目標(biāo)狀態(tài)演化模型、量測(cè)模型建立目標(biāo)狀態(tài)和數(shù)據(jù)關(guān)聯(lián)的聯(lián)合建議分布函數(shù),采用粒子濾波對(duì)多個(gè)擴(kuò)展目標(biāo)狀態(tài)和數(shù)據(jù)關(guān)聯(lián)進(jìn)行聯(lián)合估計(jì),提出了非線性系統(tǒng)的多擴(kuò)展目標(biāo)跟蹤算法。在此基礎(chǔ)上,提出了順序采樣粒子濾波器來(lái)解決維數(shù)災(zāi)難的問(wèn)題。
5 結(jié)語(yǔ)
針對(duì)非線性多擴(kuò)展目標(biāo)跟蹤,本文采用粒子濾波對(duì)多擴(kuò)展目標(biāo)狀態(tài)和數(shù)據(jù)關(guān)聯(lián)進(jìn)行聯(lián)合跟蹤,提出了多擴(kuò)展目標(biāo)粒子濾波器, 解決了目標(biāo)狀態(tài)估計(jì)和數(shù)據(jù)關(guān)聯(lián)相互牽制的問(wèn)題,減小了非線性和關(guān)聯(lián)假設(shè)的不確定性帶來(lái)的估計(jì)誤差。仿真結(jié)果表明,在初始時(shí)刻、目標(biāo)出現(xiàn)時(shí)刻以及目標(biāo)相距較近時(shí)對(duì)位置跟蹤效果較好,目標(biāo)狀態(tài)演化模型與目標(biāo)實(shí)際狀態(tài)演化相差較大時(shí)位置估計(jì)精度明顯較高,而形狀估計(jì)的性能明顯優(yōu)越。本文并未對(duì)形狀的表示方式進(jìn)行研究,下一步的研究方向是在建立復(fù)雜形狀的表示和量測(cè)源模型建立的基礎(chǔ)上,研究本文算法的適用性。
參考文獻(xiàn) (References)
[1] ??? BAUM M, HANEBECK U D. Shape tracking of extended objects and group targets with starconvex RHMs [C]// Proceedings of the 14th International Conference on Information Fusion. Piscataway, NJ: IEEE, 2011: 338-345.
[2] ??? DOUCET A, GODSILL S, ANDRIEU C. On sequential Monte Carlo sampling methods for Bayesian filtering [J]. Statistics and Computing, 2000, 10(3):197-208.
[3] ??? 韓玉蘭, 朱洪艷, 韓崇昭. 采用隨機(jī)矩陣的多擴(kuò)展目標(biāo)濾波器[J]. 西安交通大學(xué)學(xué)報(bào), 2015, 49(7): 98-104. (HAN Y L, ZHU H Y, HAN C Z. A multitarget filter based on random matrix[J]. Journal of Xian Jiaotong University, 2015, 49(7): 98-104.)
[4] ??? MAHLER R. PHD filters for nonstandard targets,I: extended targets[C]// Proceedings of the 12th International Conference on Information Fusion. Piscataway, NJ: IEEE, 2009: 915-921.
[5] ??? ORGUNER U, LUNDQUIST C, GRANSTROM K. Extended target tracking with a cardinalized probability hypothesis density filter [C]// Proceedings of the 14th International Conference on Information Fusion. Piscataway, NJ: IEEE, 2011: 1-8.
[6] ??? GRANSTROM K,LUNDQUIST C,ORGUNER U. Extended target tracking using a Gaussian mixture PHD filter [J]. IEEE Transactions on Aerospace and Electronic Systems, 2012, 48(4): 3268-3286.
[7] ??? GRANSTROM K, ORGUNER U. A PHD filter for tracking multiple extended targets using random matrices [J]. IEEE Transactions on Signal Processing, 2012, 60(11): 5657-5671.
[8] ??? HIRSCHER T, SCHEEL A, REUTER S, et al. Multiple extended object tracking using Gaussian processes [C]// Proceedings of the 19th International Conference on Information Fusion. Piscataway, NJ: IEEE, 2016: 868-875.
[9] ??? OZKAN E, WAHLSTROM N, GODSILL S J. RaoBlackwellised particle filter for starconvex extended target tracking models [C]// Proceedings of the 19th International Conference on Information Fusion. Piscataway, NJ: IEEE, 2016: 1193-1199.
[10] ?? LAN J, LI X R. Extended object or group target tracking using random matrix with nonlinear measurements [C]// Proceedings of the 19th International Conference on Information Fusion. Piscataway, NJ: IEEE, 2016: 901-908.
[11] ?? 韓玉蘭, 朱洪艷, 韓崇昭. 多擴(kuò)展目標(biāo)的高斯混合概率假設(shè)密度濾波器[J].西安交通大學(xué)學(xué)報(bào), 2014, 48(4): 95-101. (HAN Y L, ZHU H Y, HAN C Z. Gaussianmixture probability hypothesis density filter for multiple extended target[J]. Journal of Xian Jiaotong University, 2014, 48(4): 95-101.)
[12] ?? 王雪, 李鴻艷, 孔云波,等. 基于星凸RHM的擴(kuò)展目標(biāo)SMCPHD濾波[J]. 計(jì)算機(jī)應(yīng)用研究, 2017, 34(7):2144-2147.(WANG X, LI H Y, KONG Y B, et al. SMCPHD filter for extended target tracking based on starconvex random hypersurface models[J]. Application Research of Computers, 2017, 34(7):2144-2147.)
[13] ?? VERMAAK J, GODSILL S J, PEREZ P. Monte Carlo filtering for multitarget tracking and data association[J]. IEEE Transactions on Aerospace and Electronic Systems, 2005, 41(1): 309-332.
[14] LIU J S. Metropolized independent sampling with comparisons to rejection sampling and importance sampling[J]. Statistics and Computing, 1996, 6(2):113-119.
[15] DOUCET A, GODSILL S, ANDRIEU C. On sequential Monte Carlo sampling methods for Bayesian filtering[J]. Statistics and Computing, 2000, 10(3):197-208.
[16] ?? GRANSTROM K, LUNDQUIST C, ORGUNER U. Tracking rectangular and elliptical extended targets using laser measurements [C]// Proceedings of the International Conference on Information Fusion. Piscataway, NJ: IEEE, 2011: 592-599.