DTransfer: extremely low cost localization irrelevant to targets and regions for activity recognition
DTransfer: extremely low cost localization irrelevant to targets and regions for activity recognition
复制标题
DTransfer：与目标和区域无关的极低成本本地化活动识别
DOI：
10.1007/s00779-018-1177-7
复制
发表时间：
2018-09
期刊：
影响因子：
--
通讯作者：
Fang Dingyi
中科院分区：
文献类型：
--
作者：
Wang Qing;Yin Xiaoyan;Tan Jun;Xing Tianzhang;Niu Jinping;Fang Dingyi
多个基于位置的无设备活动识别系统指示可以从位置系统推断与特定位置相关的一些活动，并且添加位置信息可以提高活动识别的准确性。因此，定位技术是活动识别等应用的基础。Radio-Map是一种有效的无设备定位方法。传统的指纹系统，可以提供这样的准确性是遭受人力成本在无线电地图的建设和更新。虽然在更新阶段的人力成本已被关注，初始创建所造成的较高的成本被忽略。此外，现有的系统假设由不同目标引起的RSS变化测量在任何区域都是固定分布的。这两个缺点将极大地影响Radio-Map的实用性和鲁棒性。在本文中，我们提出，DTransfer，一个非常低成本的DFL方法，本地化不同类型的目标在不同的区域。我们设计了一种基于奇异值分解（SVD）的优化低秩矩阵完备化模型来构造感知矩阵（即，radio-map），这大大减少了开销。接下来，我们采用一个严格设计的二次转移计划，以准确地定位在不同区域的不同类别的目标。最后，我们将位置信息应用到活动识别算法中，实验表明，添加位置信息的算法的准确率提高了约6%。大量的实验结果表明，DTransfer取得了令人满意的性能。
Multiple location-based device-free activity recognition systems indicate that some activities related to specific locations can be inferred from the location system and adding location information can improve the accuracy of activity recognition. Therefore, localization technology is the basis for activity recognition and other applications. Radio-Map is an effective measure in Device-free localization (DFL). Traditional fingerprint systems that can provide such accuracy are suffering from human cost in Radio-Map construction and update. Although the human cost in update phase has been paid attention, the higher costs caused by the initially created are ignored. In addition, existing systems assume that RSS change measurements caused by different targets are fixed distribution in any region. The two drawbacks will greatly affect the practicability and robustness of Radio-Map. In this paper, we propose, DTransfer, an extremely low-cost DFL approach that localize different kinds of targets in different regions. We design an optimized low-rank matrix completion model based on singular value decomposition (SVD) to construct the sensing matrix (i.e., radio-map) of the original region, which greatly reduces the overhead. Next, we employ a rigorously designed quadratic transfer scheme to accurately locate different categories of targets in different regions. Finally, we apply the location information to the activity recognition algorithm; experiments have shown that the accuracy of the algorithm for adding location information is increased by approximately 6%. Extensive experimental results illustrate that DTransfer achieves delightful performance.
登录
查看更多内容
DOI：
10.1109/tpds.2012.134
发表时间：
2011-03
期刊：
2011 IEEE International Conference on Pervasive Computing and Communications (PerCom)
影响因子：
--
作者：
Dian Zhang;Yunhuai Liu;Xiaonan Guo;L. Ni
通讯作者：
Dian Zhang;Yunhuai Liu;Xiaonan Guo;L. Ni
DOI：
10.1109/infocom.2018.8485958
发表时间：
2018-04
期刊：
IEEE INFOCOM 2018 - IEEE Conference on Computer Communications
影响因子：
--
作者：
Yang Liu;Zhenjiang Li
通讯作者：
Yang Liu;Zhenjiang Li
影响因子：
22.7
作者：
Candes, Emmanuel;Recht, Benjamin
通讯作者：
Recht, Benjamin
DOI：
--
发表时间：
2008-07
期刊：
--
影响因子：
--
作者：
Sinno Jialin Pan;Dou Shen;Qiang Yang;James T. Kwok
通讯作者：
Sinno Jialin Pan;Dou Shen;Qiang Yang;James T. Kwok
DOI：
10.1145/2370216.2370269
发表时间：
2012-09
期刊：
Proceedings of the 2012 ACM Conference on Ubiquitous Computing
影响因子：
--
作者：
K. Yatani;K. Truong
通讯作者：
K. Yatani;K. Truong