摘要
为了解决个人隐私的保护、受光照条件和障碍物等因素的影响,设计了一种基于屋顶二值红外传感器网络的人体行为识别系统;系统采用STM32芯片和ZigBee协调器建立ZigBee网络;当实验者在安装在屋顶的20个互连的AMN31111红外传感器下面做出一系列行为时,传感器对其进行二值数据采集,ZigBee协调器将数据通过串口发送到PC机上实时动态显示,通过Keil 5系统软件对采集的二值数据以txt文本格式存储;提出了像素值法对人体进行定位和BP神经网络算法在模拟的居家环境中对7种不同人物行为进行识别;实验结果表明:该系统实现了人体多种行为的识别,其识别率为84.7%,4名实验者得到平均识别精度相比固定在居家电器传感器设备要高4.7%左右,并且该系统采集精度高、性能稳定、可靠性高、成本低、功耗低,解决了一些目前人体行为识别监测系统存在的问题。
To preserve individual privacy and to avoid the effects of light condition and obstacles, this paper designed a human activities recognition system based on binary ceiling infrared sensor network. This system uses STM32 and ZigBee coordinator to build a ZigBee net- work. When a target does some activities under the interconnection of twenty AMN31111 infrared sensors, binary values will be collected by sensors. Then ZigBee coordinator will send this data to PC real time dynamics through serial port and these values will be stored in txt format by the software of Keil 5. We proposed the method of pixel value to estimate the locations of the target. In a home environment, we used BP neural network algorithm to recognize 7 different activities of a target. The experimental results showed that this system achieved a performance of a single target＇ s activities recognition rate of 84. 7%. The average recognition rate obtained by 4 targets was about 4. 7% higher than those sensor devices attached to home appliances. The system is with high precision, stable performance, high reliability, low cost and low power consumption. So it solves some current existing problems in human behavior recognition system.
出处
《计算机测量与控制》
2017年第1期163-166,共4页
Computer Measurement &Control
基金
国家自然科学基金项目(61501076)
