自适应模态选择的鲁棒目标跟踪方法
发布时间：2019-03-24 14:33
【摘要】：在多模态跟踪过程中,为了避免低质量模态的噪声影响和提高跟踪方法的效率,提出了一种基于自适应模态选择的目标跟踪方法,能够选择较好的模态进行跟踪。具体地,对于每个模态,使用聚类方法将目标区域及其周围背景区域各聚为若干个子类,然后通过它们子类之间的特征差异衡量目标和周围背景的判别性(即模态质量),选择判别性最大的模态对目标使用相关性滤波算法进行跟踪。同时,为了维持各个模态的目标模型的有效性,提出了一种双阈值策略更新选择和未被选择模态的跟踪模型。在7组热红外和可见光视频对上进行了实验,验证了该方法的有效性,且跟踪速度达到141 f/s。
[Abstract]:In order to avoid the influence of low-quality mode noise and improve the efficiency of tracking method, a target tracking method based on adaptive mode selection is proposed in the multi-mode tracking process, which can select better modes for tracking. Specifically, for each modal, clustering methods are used to cluster the target region and its surrounding background regions into several sub-categories, and then measure the discriminance (i.e. modal quality) of the target and surrounding background through the characteristic differences between their subcategories. The most discriminant mode is selected to track the target using correlation filtering algorithm. At the same time, in order to maintain the effectiveness of the target model of each mode, a tracking model with double threshold strategy update selection and unselected mode is proposed. Experiments on 7 groups of thermal infrared and visible video pairs show that the proposed method is effective and the tracking speed is up to 141 fxs.
【作者单位】： 安徽大学计算机科学与技术学院;
【基金】：“十二五”科技部支撑计划项目No.2015BAK24B01~~
【分类号】：TN713
本文编号：2446412
[Abstract]:In order to avoid the influence of low-quality mode noise and improve the efficiency of tracking method, a target tracking method based on adaptive mode selection is proposed in the multi-mode tracking process, which can select better modes for tracking. Specifically, for each modal, clustering methods are used to cluster the target region and its surrounding background regions into several sub-categories, and then measure the discriminance (i.e. modal quality) of the target and surrounding background through the characteristic differences between their subcategories. The most discriminant mode is selected to track the target using correlation filtering algorithm. At the same time, in order to maintain the effectiveness of the target model of each mode, a tracking model with double threshold strategy update selection and unselected mode is proposed. Experiments on 7 groups of thermal infrared and visible video pairs show that the proposed method is effective and the tracking speed is up to 141 fxs.
【作者单位】： 安徽大学计算机科学与技术学院;
【基金】：“十二五”科技部支撑计划项目No.2015BAK24B01~~
【分类号】：TN713
【相似文献】
相关期刊论文 前4条
1 李波睿;慕春棣;白天明;柳志娟;;一种混合的扩展目标跟踪方法[J];航空学报;2014年05期
2 ;红外成象制导与光电对抗研讨会会议部分论文文摘(下)[J];红外与激光技术;1992年03期
3 宋新;王鲁平;王平;沈振康;;基于改进均值位移的红外目标跟踪方法[J];红外与毫米波学报;2007年06期
4 ;[J];;年期
相关博士学位论文 前1条
1 李波;基于随机有限集理论的VTS目标跟踪方法研究[D];大连海事大学;2015年
相关硕士学位论文 前2条
1 刘远;基于形状估计的随机集多扩展目标跟踪方法研究[D];西安电子科技大学;2015年
2 陈雅珍;基于随机集的扩展目标跟踪方法研究[D];西安电子科技大学;2014年
，本文编号：2446412
本文链接：https://www.wllwen.com/kejilunwen/dianzigongchenglunwen/2446412.html
