基于Beowulf机群中改进粒子滤波的3D人体运动跟踪  

Three-dimension human motion tracking based improved particle filter on Beowulf cluster system

作　　者：李敏[1,2] 宋曰聪[1] 吴斌[3] 彭保[3] 

机构地区：[1]绵阳师范学院数学与计算机科学学院,四川绵阳621000 [2]中国工程物理研究院五所,四川绵阳621000 [3]西南科技大学信息工程学院,四川绵阳621000

出　　处：《计算机工程与应用》2015年第14期17-22,77,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61001125);四川省教育厅重点项目(No.14ZA0257);绵阳师范学院重点项目(No.2011C05)

摘　　要：针对标准的粒子滤波算法在视频三维人体运动跟踪中存在的计算量巨大、粒子退化、跟踪失效而无法同时满足跟踪精度和跟踪实时性要求的问题,提出了基于Beowulf机群中改进的粒子滤波新算法。新算法通过三维人体模型参数的自动初始化、粒子数目和模板的调整来实现跟踪失效的自动恢复,基于任务动态分配策略、低开销通信策略设计的Beowulf机群中的迁移式粒子滤波并行算法克服了粒子退化问题和提高了计算速度。实验结果显示:新方法有效地减轻了粒子退化和跟踪失效问题,降低了计算时间,提高了跟踪精度,能够同时满足三维人体运动跟踪精度和实时性的要求。According to the problem that the standard particle filter tracking algorithm in video 3D human motion tracking cannot meet tracking accuracy and real time tracking at the same time for its intensive computation and particle degeneracy and tracking failure. A novel improvement particle filter algorithm is proposed based on Beowulf cluster system parallel computing. The new algorithm can realize automatic recovery from tracking failure by automatic initialization of 3D human body model parameters and adjustment of particle amount and template. The migration particle filter parallel algorithm which based on task dynamic allocation and low consumption communication strategy in Beowulf cluster system can overcome particle degeneracy problem and improve computation speed. The experimental result shows that particle degeneracy and tracking failure problems have been alleviated effectively, the computing time has been reduced, the tracking precision has been improved, and the new way can meet the need of tracking accuracy and real time tracking at the same time.

关 键 词：Beowulf机群系统 三维人体运动跟踪 迁移式粒子滤波并行算法 模板与粒子数目调整 人体模型参数自动初始化 跟踪失效的自动恢复方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]