A Vision-Based Posture Monitoring System for the Elderly Using Intelligent Fall Detection Technique
A Vision-Based Posture Monitoring System for the Elderly Using Intelligent Fall Detection Technique
复制标题
DOI：
10.1007/978-3-030-04173-1_11
复制
发表时间：
2019-01-01
期刊：
影响因子：
--
通讯作者：
Padmavathi, S.
中科院分区：
文献类型：
--
作者：
Ramanujam, E.;Padmavathi, S.
老年人监测系统是照顾独居老人和残疾人的主要应用。跌倒是老年人监测系统中需要检测到的首要因素，以避免严重伤害甚至死亡。检测系统通常使用环境传感器、可穿戴传感器和基于视觉的技术。对于基于传感器的设备，老年人需要佩戴检测设备，但是，他们经常忘记佩戴这些设备或不正确佩戴它们。此外，传感器需要定期充电和维护。此外，需要在所有房间安装环境传感器，以覆盖整个驱动过程。另外的困难是它们的电路很复杂，而且对温度很敏感。基于视觉的设备是唯一可行的解决方案，可以取代上述传感器。此外，基于视觉的实现成本要低得多，相关设备在活动识别方面也优于可穿戴设备。就像环境传感器一样，摄像头也可以安装在所有房间；与环境传感器相比，这些传感器的成本和维护成本更低。本章提出了一种基于视觉的姿势监测系统，该系统使用连接到数字视频录像机的红外摄像机和跌倒检测机制来对跌倒进行分类。在本章中，我们通过用镭反光胶带（红色）制作的特别设计的服装来观察老年人的行为，用于姿势识别。所提出的跌落检测技术包括多个操作模块，如图像分割、重新缩放和分类。红外摄像机观察老人的动作，并将信号传输到数字录像机。数字录像机只从信号中捕捉动作帧。使用图像分割将运动图像分割为红色带，并进一步使用k-最近邻和决策树分类器重新缩放以获得更好的分类。研究人员对10名不同的受试者进行了测试，以确定在仰卧、坐着、坐着伸直膝盖和站立等不同运动中跌倒的情况。我们已经证明，使用k近邻分类器提出的模型的检测率为94%。
Elderly monitoring systems are the major applications of care for elderly and the disabled who live alone. Falls are the leading factor to be detected in the elderly monitoring system to avoid serious injuries and even death. The detection systems often use ambient sensors, wearable sensor, and vision-based technologies. In case of sensor-based devices, the elderly are required to wear the detection devices, however, quite often, they forget to wear these or do not wear them correctly. Moreover, the sensors need to be charged and maintained regularly. Also, the ambient sensors need to be installed in all the rooms to cover the whole actuation. The additional difficulty is that they are complex in circuitry and sensitive to temperature. Vision-based devices are the only plausible solution that can replace the aforementioned sensors. Besides, the cost of vision-based implementation is much lower and related devices are better than wearable devices in activity recognition. Much like Ambient sensors, cameras can also be installed in all the rooms; the cost and maintenance of these are less as compared to ambient sensors. This chapter proposes a vision-based posture monitoring system using infrared cameras connected to a digital video recorder and a fall detection mechanism to classify the falls. In the chapter, we observe the behavior of the elderly through the specially designed clothing fabricated with retroreflective radium tape (red in color) for posture identification. The proposed fall detection technique comprises various modules of operations such as image segmentation, rescaling, and classification. The infrared cameras observe the movement of the elderly people and signals are transmitted to a digital video recorder. The digital video recorder snaps only the motion frames from the signal. The motion images are segmented to red band using image segmentation and further rescaled for better classification using k-Nearest Neighbor and decision tree classifiers. The tests have been conducted on 10 different subjects to identify the falls during various motions such as supine, sitting, sitting with knee extension, and standing. We have shown a detection rate of 94% for the proposed model with k-nearest neighbor classifier.