摘要
针对非刚体目标追踪算法追踪精度与计算效率较低的问题,提出了一种基于混合光流的非刚体目标追踪算法。首先,采用Lucas-Kanade方法计算局部区域的光流,采用Horn-Schunck方法计算全局区域的光流,局部光流与全局光流具有互补的特点,因此结合两种方法可获得理想的结果,并且设计了基于混合光流法与Gabor特征的非刚体目标轮廓模型;然后,基于光流的特征计算光流的流域,采用高斯混合模型分析光流的流域,从而实现运动目标的检测;最终,采用分类器对感兴趣区域进行快速地分类,动态地更新每个视频帧的目标轮廓。基于多个公开的视频数据集进行了实验分析,结果显示本算法能够有效地完成对非刚体目标的追踪,并且具有较高的追踪准确性。
Aiming at the problem of low tracking accuracy and low computational efficiency of non-rigid target tracking algorithm,a non-rigid target tracking algorithm based on mixed optical flows is proposed.Firstly,the Lucas-Kanade method is adopted to compute the optical flow of local region,Horn-Schunck method is adopted to compute the optical flow of global region.Local optical flow and global optical flow have complementary shortcomings and advantages,so that an ideal result is realized by two methods combination.At the same time,a non-rigid target contour model based on optical flow and Gabor feature is proposed;then,flow region of optical flow is computed based on the optical flow feature,and the flow region of optical flow is analyzed by Gaussian mixture model,in order to realize moving target tracking;finally,interested regions are classified rapidly by classification machine,and the target contour of each video frame is updated dynamically.Experimental results based on a few public video datasets show that the proposed algorithm realizes non-rigid target tracking effectively,and it has a high tracking accuracy.
作者
谭继安
林德丰
梁建胜
TAN Ji-an;LIN De-feng;LIANG Jian-sheng(Dongguan Polytechnic,Dongguan 523808,China;School of Information Technology in Education,South China Normal University,Guangzhou 510631,China)
出处
《控制工程》
CSCD
北大核心
2020年第5期848-854,共7页
Control Engineering of China
基金
广东省高职教育教学改革研究与实践项目(GDJG2019006)
东莞市标准化研究项目“基于物联大数据的学前教育实训室内环境风险管理及标准研究”。
关键词
光流法
目标追踪
视频处理
分类器
目标轮廓
高斯混合模型
Optical flow method
target tracking
video processing
classification machine
target contour
Gaussian mixture model
