摘要
传统的基于粒子滤波的目标跟踪中,通常待跟踪的目标或者在视频初始帧中由人工指定,或者需要对背景进行建模后由背景差方法进行分割,受人类视觉感知机制的启发,提出一种基于尺度不变特征转换(scale invariant feature transform,SIFT)流的动态视觉显著性算法,结合颜色、亮度以及方向等静态特征以实现对感兴趣目标的自动定位;通过融合静态显著性与动态显著性形成总显著图,并选择显著性最高的目标作为待跟踪的感兴趣目标;通过提取目标区域的颜色、梯度及旋转不变局部二进模式(local binary pattern,LBP)纹理等特征构建目标模板,采用粒子滤波器对目标进行跟踪。结果表明,该算法能够在一定程度上模拟人类动态视觉注意过程,有效地检测感兴趣的目标并进行稳定鲁棒的跟踪。
When the tracking moving object in video image sequences by particle filter, usually the object to be tracked was manually selected in the first video frame or segmented by background subtraction. Inspired by the human visual mechanism, an algorithm which can locate the object of interest automatically based on dynamic visual saliency modeled by scale invariant feature transform （SIFT） flow was proposed. To detect the object of interest, some other static features such as color, intensity, and orientation were also extracted. A saliency map was developed by combining both the static saliency and the dynamic saliency, and the most salient object was selected as the object of interest. This method was further applied to object tracking with particle filter. The object template was constructed by fusing the color, gradient, and local binary pattern （LBP） texture features. The results show that the proposed method can simulate the human＇ s dynamic attention process to some extent when an object starts to move in a scene and track the object of interest robustly.
作者
李蕙
王延江
LI Hui WANG Yanjiang(College of Information and Control Engineering,China University of Petroleum （East China） ,Qingdao 266580, Chin)
出处
《济南大学学报（自然科学版）》
CAS
北大核心
2016年第5期368-376,共9页
Journal of University of Jinan(Science and Technology)
基金
国家自然科学基金(61271407)
关键词
视觉显著性
运动显著性
尺度不变特征转换流
粒子滤波
局部二进模式
目标跟踪
visual saliency
motion saliency
scale invariant feature transform flow
particle filter
local binary pattern
object tracking
