智能住宅老年人异常行为

监测系统设计与研究

陈 坤 辉

摘 要

随着世界经济持续发展，人口老龄化问题日益突出。为此，我国提出“9073”养老模式，即在全市户籍老年人中，90%由家庭自我照顾，7%享受社区居家养老服务，3%享受机构养老服务。然而现有的居家养老服务从业者人数和专业性远远达不到社会需求。随着人工智能、物联网技术的发展，智能住宅技术能够为独居老年人提供一系列的生活辅助，跟踪住户健康状况、实现居民异常行为识别，能有效缓解这一难题。然而，智能住宅在实际研发和应用中还存在以下几个问题：（1）当前智能住宅主要应用于智能家居、环境监测和智能安防等，关注居民的身体健康状况、居家异常行为的应用较少；（2）智能住宅传感系统的风险意识不强，在收集住户健康信息的同时，也要注重隐私的保护；（3）对老年人异常行为的监测，目前主要采用手环、RFID标签等方式，对用户异常行为识别的精度不高，误报情况较多。

为解决以上问题，本研究提出了一种新的居家异常行为检测思路，针对老年人居家异常行为，提供检测和预警。主要工作如下：（1）梳理老年人常见异常行为、发生原因，构建了异常俯身和跌倒行为模型。从计算机视觉、可穿戴设备和室内定位传感器的角度对典型的异常行为检测方法进行梳理；（2）提出基于普通相机获取人体灰度图，并通过深度学习的技术和欧式变换等方法获得人体骨架图及其三维坐标的方法；（3）构建了一套异常行为监测系统。结合老年人异常俯身和跌倒行为模型提取人体骨架的关键特征点，使用SVM机器学习的方法对数据进行分类和识别，实现了对老年人居家异常行为的识别，实验准确率达97%。

本文所构建的方法及异常行为监测系统能够大大提高居家老年人异常行为感知效果，用于面向老年人的智能住宅的构建，能有效减缓有关政府部门和家庭的养老压力，具有重要的研究和应用价值。

关键词：智能住宅；老年人异常行为；行为异常检测；计算机视觉；支持向量机

Abstract

As the world economy continues to develop, the problem of population aging is becoming increasingly prominent. To this end, China proposes the “9073” pension model, that is, 90% of the elderly are self-care by the family, 7% enjoy the community home care service, and 3% enjoy the institutional pension service. However, the number of existing home care service practitioners and professionalism is far away, not meeting the requirements. With the development of artificial intelligence and Internet of Things technology, smart home technology can provide a series of life support for elderly people living alone, track the health status of residents, and realize the identification of abnormal behavior of residents, which can effectively alleviate this problem. However, the following problems exist in the actual development and application of smart home: (1) The current smart home is mainly used in smart furniture, environmental monitoring and intelligent security, paying no attention to the health status of residents and the application of abnormal behavior detection at home. (2) The risk awareness of smart residential sensing systems is not strong. At the same time, we must also pay attention to the protection of privacy while collecting household health information. (3) For the monitoring of abnormal behaviors of the elderly, the current use of wristbands, RFID tags and the accuracy of the user's abnormal behavior recognition is not high.

In order to solve the above problems, this study proposes a new abnormal behavior detection idea, providing detection and early warning for the abnormal behavior of elderly at home. The main work is as follows: (1) Combing the common abnormal behaviors and causes of the elderly, and constructing a model of abnormal leaning and falling behavior. Typical abnormal behavior detection methods are combed from the perspective of computer vision, wearable devices and indoor positioning sensors. (2) Propose a method for obtaining the human body gray map based on the ordinary camera and obtaining the human skeleton map and its three-dimensional coordinates by deep learning techniques and Euclidean transformation (3) A set of abnormal behavior monitoring system was constructed. Combining the abnormality of the elderly and the fall behavior model to extract the key features of the human skeleton, the SVM machine learning method is used to classify and identify the data, and the recognition of the abnormal behavior of the elderly is realized. The experimental accuracy rate is 97%.

The method and abnormal behavior monitoring system constructed in this study can greatly improve the abnormal behavior perception effect of the elderly in the home. It is used for the construction of intelligent housing for the elderly, which can effectively alleviate the pension pressure of relevant government departments and families. It has important research and application. value.

Keywords：Smart homes; Abnormal behavior in the elderly；Abnormal behavior detection; Computer vision; SVM