摘要
针对空中目标的跟踪环境提出一种边缘幅值分布的相关跟踪算法,利用目标的边缘幅值分布作为目标的特征模板,通过求取当前帧中目标特征与目标特征模板相关系数的最优解来进行目标跟踪。与采用对称核函数的均值迁移目标跟踪算法相比,采用目标边缘点作为核函数中的样本点,参与计算的样本点为核窗口中样本的5%～10%,使图像处理速度达到了50帧/s以上,满足了实时跟踪的要求。在跟踪过程中,以目标相邻帧间特征向量的Bhattacharyya相关系数作为目标特征模板的更新判据,实验中相邻帧间目标特征向量的Bhattacharyya系数保持在0.95～1.0,满足模板实时更新的要求,为稳定跟踪提供了保障。
A real-time correlation tracking algorithm, based on edge amplitude distribution, is presented to solve the aerial object tracking problem. We use the edge amplitude distribution histogram to express the target and track the target by computing the best correlation between the object model and the current object. The traditional mean shift algorithm requires a symmetric kernel, such as a circle or a rectangle, and assumes that the target＇s scale is symmetric during the course of tracking. The pixels are used on the edge as the samples, about 5% to 10% of the total pixels in the kernel. The object model is updated by computing the Bhattacharyya coefficient. The experimental results show that the Bhattacharyya coefficient during the course of tracking keeps between 0.95 and 1.0, and algorithm runs well at 50 fps.
出处
《光电工程》
CAS
CSCD
北大核心
2009年第5期28-33,共6页
Opto-Electronic Engineering
基金
"985"工程学科建设投资项目(107008200400020)
