Abstract

Wireless device-free human sensing is an emerging technique of Internet of Things, which holds great potential for ubiquitous location-based services and human-interaction applications. Although existing studies can detect human appearance, they still neglect to further identify whether a user is approaching a sensor or not, which is critical for fine-grained recognition of human behaviors. In this paper, we first conduct comprehensive experiments to measure relationships between signal fading and human positions. The experimental results show that signal fading stepwise changes with different distances of the human to a sensor. Moreover, the signal fading is worse when the human is located closer to an antenna of the sensor. Motivated by these observations, we propose NSee, a novel system for device-free near-field human sensing without site-survey fingerprints. Specifically, we cluster signal fading features of different antennas by a Gaussian mixture model, and further propose a cluster identification algorithm to identify clusters in correspondence to different near-field subareas of human appearance. Based on cluster characteristics, NSee can recognize near-field human presence with online sensing. We implement a prototype of NSee system based on a commercial WiFi card with multiple antennas. Extensive experimental results illustrate that the proposed system can achieve an averaged accuracy of 90% in device-free near-field human recognition.

中文翻译：

---

使用WiFi信号的无设备近场人体感应

摘要

无需无线设备的人感测是物联网的新兴技术，它对于无处不在的基于位置的服务和人机交互应用具有巨大的潜力。尽管现有研究可以检测出人类的外观，但他们仍然忽略了进一步识别用户是否正在接近传感器的情况，这对于细粒度地识别人类行为至关重要。在本文中，我们首先进行全面的实验以测量信号衰减与人体位置之间的关系。实验结果表明，信号衰落随着人类到传感器的不同距离而逐步变化。此外，当人靠近传感器的天线时，信号衰落更严重。基于这些观察，我们提出了NSee，一种无需站点调查指纹的无需设备的近场人体感应的新颖系统。具体地，我们通过高斯混合模型对不同天线的信号衰落特征进行聚类，并进一步提出了一种聚类识别算法，以识别与人类外观的不同近场子区域相对应的聚类。基于群集的特征，NSee可以通过在线感应识别近场人类的存在。我们基于带有多个天线的商用WiFi卡实现NSee系统的原型。大量的实验结果表明，该系统在无设备近场人类识别中可以达到90％的平均准确度。