基于深度稀疏滤波的目标跟踪
发布时间：2018-11-20 16:35
【摘要】：为了在复杂环境下更好地区分被跟踪目标和背景,设计了一种基于2l范数归一化和1l范数最小化的深度稀疏滤波模型,通过深度学习获取样本稀疏特征并对其进行分类,将该模型和粒子滤波框架结合,提出一种目标跟踪算法.首先使用离线训练集对深度稀疏滤波模型进行逐层无监督预训练得到权值参数的初始值,然后在跟踪过程中利用标签样本对模型在线更新,通过有监督微调优化其权值参数使得模型能够更好地适应目标外观变化,最后利用训练好的深度稀疏滤波模型对经由粒子滤波算法给出的候选区域进行观测,以确定跟踪目标.在不同视频序列中的实验表明,文中算法在复杂条件下具有良好的跟踪精度和鲁棒性.
[Abstract]:In order to distinguish the target from the background better in complex environment, a deep sparse filtering model based on 2l norm normalization and 1L norm minimization is designed. The sparse feature of the sample is obtained and classified by depth learning. A target tracking algorithm is proposed by combining the model with the particle filter framework. Firstly, the initial value of weight parameters is obtained by using off-line training set to pre-train the depth sparse filter model layer by layer, and then the model is updated online by label samples during the tracking process. The weight parameters are optimized by supervised fine-tuning so that the model can better adapt to the appearance changes of the target. Finally, the candidate regions given by particle filter algorithm are observed by using the trained depth sparse filter model to determine the target tracking. Experiments in different video sequences show that the proposed algorithm has good tracking accuracy and robustness under complex conditions.
【作者单位】： 闽江学院物理学与电子信息工程系;
【基金】：国家自然科学基金(51277091) 中国博士后科学基金(2013T60637) 福建省中青年教师教育科研项目(A15415) 福州市科技计划重点项目(2013-G-86)
【分类号】：TP391.41
本文编号：2345409
[Abstract]:In order to distinguish the target from the background better in complex environment, a deep sparse filtering model based on 2l norm normalization and 1L norm minimization is designed. The sparse feature of the sample is obtained and classified by depth learning. A target tracking algorithm is proposed by combining the model with the particle filter framework. Firstly, the initial value of weight parameters is obtained by using off-line training set to pre-train the depth sparse filter model layer by layer, and then the model is updated online by label samples during the tracking process. The weight parameters are optimized by supervised fine-tuning so that the model can better adapt to the appearance changes of the target. Finally, the candidate regions given by particle filter algorithm are observed by using the trained depth sparse filter model to determine the target tracking. Experiments in different video sequences show that the proposed algorithm has good tracking accuracy and robustness under complex conditions.
【作者单位】： 闽江学院物理学与电子信息工程系;
【基金】：国家自然科学基金(51277091) 中国博士后科学基金(2013T60637) 福建省中青年教师教育科研项目(A15415) 福州市科技计划重点项目(2013-G-86)
【分类号】：TP391.41
【相似文献】
相关期刊论文 前4条
1 高颖慧;沈振康;;量子衍生窗口滤波模型[J];计算机应用研究;2009年11期
2 余学祥,吕伟才;GPS监测网动态数据处理抗差Kalman滤波模型[J];中国矿业大学学报;2000年06期
3 强成虎;李铮;郑铮;王希东;;高精度GPS时间信息的研究与应用[J];国外电子测量技术;2007年03期
4 ;[J];;年期
相关硕士学位论文 前3条
1 宓庆华;一族保边方向偏微分方程滤波模型及应用研究[D];天津大学;2014年
2 甘雨;基于时频特性分析的INS误差处理及组合导航系统滤波模型精化[D];解放军信息工程大学;2012年
3 杨娜;空间域保边滤波方法及在核级石墨三点弯曲实验中的应用[D];天津大学;2013年
，本文编号：2345409
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2345409.html
