True Presence Detection via Passive Infrared Sensor Network Using Liquid Crystal Infrared Shutters
True Presence Detection via Passive Infrared Sensor Network Using Liquid Crystal Infrared Shutters
复制标题
使用液晶红外快门通过被动红外传感器网络进行真实存在检测
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
Ya Wang
中科院分区：
文献类型：
--
作者：
Libo Wu;Ya Wang
最近，智能家居应用正在快速增长，包括但不限于照明、加热和冷却的占用相关控制。被动红外（PIR）传感器在这些应用中起着重要的作用，用于感知人体的存在和/或运动。然而，PIR传感器不能检测静止的居住者，而静止的存在占据了一天中的大部分时间。因此，产生的假阴性检测导致不舒服的光/温度波动，缩短设备的寿命，和/或能源浪费等，为了解决这个问题，我们的小组已经开发了同步低能量电子斩波PIR（SLEEPIR）传感器，集成了电子液晶（LC）红外快门与现成的PIR传感器。在这项工作中，红外快门由聚合物分散液晶（PDLC）夹在两个锗窗口，提出了周期性地斩波长波红外信号接收的PIR传感器，使静止的人仍然可以检测到由于电子快门。制作了由无线微控制器、SLEEPIR传感器和传统PIR传感器组成的传感器模块，视场为103° × 103°。然后，传感器网络由两个传感器模块的开发。本文进行了三种类型的实验：基于个人行动的，基于连续活动的，基于日常惯例的。通过使用阈值分类方法来进行检测逻辑，其中阈值从基于动作的数据集确定并应用于其他两个数据集。结果表明，对于基于活动的数据集，平均准确率达到98.96%。对于日常数据集，平均准确率为99.57%。
Recently, smart home applications are increasing fast, including but not limited to occupancy-dependent control of lighting, heating and cooling. Passive infrared (PIR) sensors play an important role in these applications to perceive the presence and/or the motion of human. However, PIR sensors are not able to detect stationary occupants while stationary presence takes up most time of the day. And thus, the resulted false negative detection leads to uncomfortable light/temperature swings, shortened equipment’s lifespan, and/or energy waste, etc. To address this issue, our group has developed Synchronized Low-Energy Electronically-chopped PIR (SLEEPIR) sensors that integrate an electronic liquid crystal (LC) infrared shutter with an off-the-shelf PIR sensor. In this work, the infrared shutter made of polymer dispersed liquid crystal (PDLC) sandwiched by two germanium windows is proposed to periodically chop the long-wave infrared signal received by the PIR sensor so that stationary human presence can still be detected due to the electronical shuttering. A sensor module is fabricated, consisting of a wireless microcontroller, a SLEEPIR sensor and a traditional PIR sensor, with a field of view of 103° × 103°. Then, a sensor network consists of two sensor modules is developed. Three types of experiments are conducted in this paper: individual action-based, continuous activity-based, and daily routine-based. The detection logic is made by using the threshold value classification method, where the threshold values are determined from the action-based dataset and applied to the other two datasets. The results show that for activity-based dataset, the average accuracy reached 98.96%. For daily routine-based dataset, the average accuracy is 99.57%.