基于Mean_Shift和粒子滤波的两步多目标跟踪方法.pdfVIP

22 1 () Vol. 22 No. 1 20102 Journal ofChongqing University ofPosts and Teleco unications(Natural ScienceEdition) Feb. 2010 Mean-Shift 李红波, 曾德龙, 吴 渝 ( , 400065) : 针对M ean-Sh ift跟踪算法 易跟踪丢失以及粒子滤波跟踪算法计算量大等问题, 提出了一种两步多目标跟 踪算法利用 M ean-Sh ift进行第一步跟踪得到候选目标,在跟踪不准的情况下再利用粒子滤波进行后续的跟踪结 果修正实验结果表明两步跟踪算法既能保持M ean-Sh ift跟踪的实时性, 也能维持粒子滤波跟踪算法的鲁棒性,有 很高的实用性 : M ean-Sh ift; 粒子滤波; 两步跟踪; 目标跟踪 : TP391 : A : 1673-825X( 2010) 01-0112-06 A two-step ultiple targets tracking algorith based on Mean-Shiftand particle filter IHong-bo, ZENG De- long, WU Yu ( Institute ofA rtificial Intelligence, Chongqing University of Posts and Telecommunications, Chongqing 400065, P. R. China) Abstract: SinceM ean-Sh ift (M S) track ing algorithm awl ays loses tracking objects and the particle filter track ing algorithm costs huge compu tation, a novel tw o-step mu ltiple targets tracking algorithm was proposed. F irstly, a candidate objectw as gotten byM ean-Shift (MS) tracking algorithm. Then, the track ing resu ltw ou ld be verified by particle filter techniquewhen the object couldntbe traced exactly. Expermi ental resu lts show that th is approach can m aintain the efficiency ofM S algo- rithm and the powerfu l ab ility of particle filter techn ique, so it is of h igh practicab ility. Key words: M ean-Shift (MS) ; particle filter ( PF); two-step track ing; object track ing , , 0 ; ( object tracking) , , (), , ,