基于简化的ISPF算法与视听信息融合的目标跟踪  

Target tracking based on audio-visual information fusion and simplified ISPF algorithm

作　　者：曹洁[1] 魏建勇[1] 刘宗礼[1] 王进花[1] 

机构地区：[1]兰州理工大学计算机与通信学院,兰州730050

出　　处：《仪器仪表学报》2011年第1期76-80,共5页Chinese Journal of Scientific Instrument

基　　金：甘肃省自然科学基金(1010RJZA046);甘肃省财政厅项目(0914ZTB148)资助

摘　　要：提出了一种适用于视听融合的简化型重要性函数粒子滤波器(ISPF),该算法充分利用视听媒体间的天然时空相关性,基于视听信息的融合实现了特定场景中对说话人的跟踪。将声源定位结果转换为一个具有特殊权重的粒子,以图像彩色信息模板匹配算法作为视觉底层跟踪,通过具有特殊权重的粒子来限制和引导视觉跟踪结果。实验结果表明,文中算法与基于单一信息的目标跟踪相比平均误差降低了约三倍,具有明显的优越性。A simplified ISPF（Incremental SPF） for audio-visual fusion is presented here.The algorithm makes full use of natural spatial and temporal correlation between audio-visual media,and speaker tracking in specific scene can be achieved based on audio-visual information fusion.The algorithm transforms sound source location result into a particle with special weight.Then image color information template matching algorithm is used as underlying visual tracking.The visual tracking result is restricted and guided through the special weighted particle.Experimental result shows that this algorithm has obvious advantage;and compared with target tracking method with single information,this algorithm decreases the average error by about 3 times.

关 键 词：目标跟踪 声源定位 视听信息融合 模版匹配 ISPF 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]