摘要
提出了一种基于LBP(Local Binary Pattern)纹理特征的红外成像目标跟踪方法,将LBP纹理特征集成到了核跟踪方法中.根据目标各区域对背景的区分能力不同,提出了目标各区域置信度的评价方法,用基于区域置信度及空间距离核加权的LBP特征概率密度函数,构造了目标及候选目标的特征模型.通过相似性度量,利用均值漂移方法实现了基于纹理特征的红外成像目标跟踪.实验结果验证了该算法在红外成像目标跟踪中较基于灰度的均值漂移跟踪算法更为鲁棒.
An IR imaging target tracking method is proposed,which integrates LISP (Local Binary Pattern) texture feature with kernel-based tracking. The confidence of each target region is calculated according to its discriminative power between the target region and local background. The target and candidate models are characterized by the kernel density estimation of LISP texture feature weighted by region confidence and distance. The similarity function between target and candidate is optimized by mean shift algorithm to realize real-time tracking. The effectiveness of the method is demonstrated by several real sequences testing.
出处
《光子学报》
EI
CAS
CSCD
北大核心
2007年第11期2163-2167,共5页
Acta Photonica Sinica
基金
国家自然科学基金重点项目(60634030)
国家自然科学基金( 60602056 )
高等学校博士学科点专项基金(20060699032)
杭州航空科学基金(2007ZC53037)资助