共享存储并行多目标跟踪.pdfVIP

第 28卷第 9 期 计算机应用 Vo l. 28 No. 9 2008年 9 月 　 Compu ter App lication s 　 Sep. 2008 文章编号 : 100 1 - 9081 (2008) 09 - 2303 - 03 共享存储并行多目标跟踪 王孝刚 ,吴晓娟 ,周 　鑫 ,张小燕 (山东大学 信息科学与工程学院 ,济南 250100) (wond@m ail. sdu. edu. cn) 摘 　要 :高度的运算复杂性制约了粒子滤波在实际的多 目标视频跟踪系统中的应用 。为克服性能瓶颈 ,探索了 一种基于 Op enM P共享存储并行编程模型的粗粒度并行多目标跟踪系统的实现方法 。在共享变量中维护被跟踪 目标 的列表 ,每一个 目标用一个独立的粒子滤波器进行跟踪 。根据处理单元的数 目确定线程数量和每个线程跟踪的目标 数量 。与对应的串行版本相比 ,该并行系统将可实时跟踪的目标数 目由 2个增加到了 8个 ,具有更大的实用价值 。 关键词 :多目标跟踪 ;粒子滤波 ;共享存储并行编程 中图分类号 : TP39 1. 4 1; TP242. 6 + 2　　文献标志码 : A Sharedm em ory para llel m ultitarget track ing WAN G X iaogang, WU X iaojuan , ZHOU X in , ZHAN G X iaoyan (S chool of Inf orm a tion S cience and E ng ineering, S handong Un iversity, J inan S handong 250 100, Ch ina) A b stract: The app lication of p article filtering in real videobased m u ltitarget track ing system s is lim ited becau se of its h igh compu tational comp lexity. To overcom e the efficiency bottleneck, a coarsegrained p arallel m u ltitarget track ing imp lem entation ba sed on the Op enM Psp ecified sharedm emory p arallel p rogramm ing model was exp lored. A list of tracked targets wa s m ain tained in a shared variab le, and each target was tracked by an indep endent p article filter. The num ber of thread s and the number of targets tracked by each thread were determ ined by the num ber of p roce ssing un its. Comp ared to its corresponding op tim ized sequential version, the p arallel imp lem entation, wh ich increa se s the num ber of targ