室内被动定位技术研究及其在行为监测中的应用

发布时间：2018-11-17 15:56

【摘要】：随着室内环境中基于位置服务的需求快速的增长,基于指纹识别的室内定位因其较高的精度引起了广泛的关注。接收信号强度指示(RSSI)作为一种常规的方案被广泛的用于位置导航系统和定位系统,但是室内复杂环境产生的多径效应导致系统的精确性得不到保障。近年来,物理层的信道状态信息(CSI)能够被更多的无线商用设备获取,它能更细粒度展现信号的特征,而且拥有更好的稳定性。本文中,提出了一种基于CSI指纹的室内被动定位算法,能更加精确的估计出目标的具体坐标位置。首先采用基于密度的聚类算法DBSCAN去除原始数据中的噪点,降低离群数据的干扰;然后使用主成分分析法(PCA)提取特征中贡献率高的项目,降低特征维度和计算复杂度;最后结合支持向量机(SVM)的回归算法建立CSI指纹与位置坐标的关联模型。同时,还将CSI指纹运用于行为监测,在入侵检测中使用SVM的二分类方法检测入侵的发生;在简单目标识别中使用SVM的多分类方法区分目标;在室内目标计数中使用基于权值的膨胀矩阵法结合SVM回归算法计算目标个数;在人群密度检测中使用动态时间归整(DTW)算法匹配最佳的人群密度。实验结果显示,本文提出的定位算法平均定位误差距离为1.37米,通过与多种定位方法对比,证明该方法在定位的精度上有明显的优势;在入侵检测中,门口入侵检测和房间有人检测的准确度分别达到98.2%和99.1%;简单目标识别中分类准确率为98.7%;室内目标计数的平均数目误差数量为0.62;人群密度的准确率为95%。实验证明本文提出的基于CSI指纹的行为监测切实可用,可实现行为的准确监测。
[Abstract]:With the rapid growth of the demand for location-based services in indoor environment, fingerprint based indoor location has attracted wide attention due to its high accuracy. As a conventional scheme, the received signal intensity indication (RSSI) is widely used in position navigation systems and positioning systems, but the accuracy of the system is not guaranteed due to the multipath effect caused by complex indoor environment. In recent years, the physical layer channel state information (CSI) can be obtained by more wireless commercial devices, it can show the characteristics of the signal more fine-grained, and has better stability. In this paper, an indoor passive location algorithm based on CSI fingerprint is proposed, which can estimate the exact position of the target more accurately. Firstly, the density based clustering algorithm (DBSCAN) is used to remove the noise in the original data to reduce the disturbance of outlier data, and then the principal component analysis (PCA) is used to extract the items with high contribution rate to reduce the feature dimension and computational complexity. Finally, combining the regression algorithm of support vector machine (SVM), the correlation model between CSI fingerprint and position coordinates is established. At the same time, the CSI fingerprint is applied to behavior monitoring, and the two-classification method of SVM is used to detect the occurrence of intrusion in intrusion detection, and the multi-classification method of SVM is used to distinguish the object in simple target recognition. In the indoor target counting, the expansion matrix method based on weight and the SVM regression algorithm are used to calculate the number of targets, and the dynamic time normalized (DTW) algorithm is used to match the optimal population density in crowd density detection. The experimental results show that the average positioning error distance of the proposed algorithm is 1.37 meters. Compared with many localization methods, this method has obvious advantages in positioning accuracy. In intrusion detection, the accuracy of door intrusion detection and room human detection is 98.2% and 99.1 respectively, the classification accuracy of simple target recognition is 98.7, the average number of errors of indoor target count is 0.62. The accuracy of crowd density was 95%. The experimental results show that the proposed CSI fingerprint based behavior monitoring is feasible and can be used to accurately monitor behavior.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN911.7

【参考文献】