结合目标预估计与Mean Shift理论的运动目标跟踪算法  

Moving object tracking algorithm based on object estimation and Mean Shift theory

作　　者：赵倩[1] 袁健全[2] 鲁新平[3] 李吉成[3] 

机构地区：[1]西北工业大学电子信息学院,陕西西安710072 [2]北京机电工程研究所．北京100074 [3]国防科学技术大学ATR实验室,湖南长沙410073

出　　处：《红外与激光工程》2010年第6期1152-1156,共5页Infrared and Laser Engineering

基　　金：装备预先研究项目

摘　　要：图像的运动包括目标、背景和平台的运动,复杂的运动关系增加了目标跟踪的难度。提出了一种有效的基于Mean Shift理论的运动目标跟踪算法。为提高算法的实时性,对Mean Shift算法的核函数进行了改进,使得加减运算替代乘方和浮点运算,大大提高了运算效率;并通过对迭代权值的改进,强化了初始模板的主要信息,提高了算法跟踪与背景相似目标的能力;采用自动更新模板的策略,克服了目标特征分布发生改变的问题;在此基础上,引入了目标预检测,提出了综合背景差分检测的运动目标跟踪算法,实验表明:该算法在目标被遮挡情况下具有较强的适应性。Image movement involves target movement,background movement and platform movement.These movement complexity increases the difficulty of target tracking.In this paper,an efficient improved algorithm based on Mean Shift theory for tracking moving object was proposed.The improvement on the kernel function was adopted to accelerate the tracking algorithm,which makes plus,subtraction replace of floating-point operations and power operations.To enhance the algorithm ability of tracking target similar to background,an improved iterative weight was proposed to strengthen the key information of initial template.To overcome the problems related to the changes in the target feature distributions,the strategy was used to automatically update the target template.Meanwhile,a moving object tracking algorithm combined with background subtraction detection was proposed.The experiment results show that the algorithm in this paper can track the moving target efficiently with object occlusion.

关 键 词：运动目标跟踪 Mean SHIFT 核函数 迭代权值 背景差分 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]