基于二阶空间直方图的双核跟踪  

Dual-kernel Tracking Approach Based on Second-order Spatiogram

作　　者：张灿龙[1,2] 唐艳平[2] 李志欣[1,2] 马海菲[1,2] 蔡冰[1,2] 

机构地区：[1]广西师范大学计算机科学与信息工程学院,桂林541004 [2]桂林电子科技大学广西信息科学实验中心,桂林541004

出　　处：《电子与信息学报》2015年第7期1660-1666,共7页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(61365009;61165009);广西自然科学基金(2014GXNSFAA118368;2012GXNSFAA053219);广西高校科技项目(2013YB027);广西信息科学实验中心经费资助课题

摘　　要：针对核跟踪算法中的背景信息弃用和空间结构丢失问题,该文提出一种基于二阶空间直方图的双核式目标跟踪算法。该算法以二阶空间直方图为目标表示模型,以相似度和对比度为目标判断准则,来建立全新的目标函数;并依据多变量泰勒展开和目标函数最大化方法,推导出双核式目标位移公式;最后使用均值漂移程序递归地获得了目标的最优位置。通过对各种条件下运动目标的跟踪验证了算法的有效性。In order to avoid the loss of background and spatial information in mean shift tracker, a dual-kernel tracking approach based on the second-order spatiogram is proposed. In the method, the second-order spatiogram is employed to represent a target, the similarity and contrast are considered simultaneously when evaluating the target candidate, and they are adaptively integrated into a novel objective function. By performing multi-variable Taylor series expansion and maximization on the objective function, a dual-kernel target location-shift formula is induced. Finally, the optimal target location is gained reeursively by applying the mean shift procedure. Experimental evaluations on several image sequences demonstrate the effectiveness of the proposed algorithm.

关 键 词：模式识别 双核跟踪 空间直方图 延森-香农散度 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]