摘要
为缓解社会老龄化压力和解决子女照顾老人的精力不足等问题,设计一种基于深度相机的高准确性和低误报性的人体跌倒检测系统。该系统使用RGB相机和红外IR相机获取标定后的老人所在环境的3D图像,并利用深度卷积神经网络提取人体的多个关节点位置,最后基于多个连续帧之间人体关节点的运动变化特征和3D场景特征相结合的方法综合判定老人是否发生跌倒行为。测试实验结果表明该系统能有效地检测到人体的跌倒行为,具有十分优良的鲁棒性。
A fall detection system using a depth camera with a high accuracy and a low false alarm rate is designed to ease the pressure of children who are lacking the energy to take care of their elderly parents and alleviate the pressure caused by the increasingly aging. In this scheme, calibrated RGB camera and infra red camera are used to obtain the 3 D image of the surroundings of the elderly, and convolutional neural network is used to extract the multijoint positions. Whether fall occurs is determined by the comprehensive consideration of the 3 D information of surrounding and the space-time characteristics of human joints between video frames. The test results show that the proposed fall detection system has a high accuracy and superior roubst for the elderly.
引文
[1]耳玉亮,段蕾蕾,叶鹏鹏,等. 2014年全国伤害监测系统老年跌倒/坠落病例特征分析[J].中华流行病学杂志, 2016, 37(1):24-28.ER Y L, DUAN L L, YE P P, et al. Epidemiologic characteristics of fallin old population:results from national injury surveillance in China,2014[J]. Chinese Journal of Epidemiology, 2016, 37(1):24-28.
[2]郑娱,鲍楠,徐礼胜,等.跌倒检测系统的研究进展[J].中国医学物理学杂志, 2014, 31(4):5071-5076.ZHENG Y, BAO N, XU L S, et al. The research progress of falldetection system[J]. Chinese Journal of Medical Physics, 2014, 31(4):5071-5076.
[3]秦晓华,段侪杰,袁克虹,等.一种老年人移动健康监护系统的研究[J].中国医学物理学杂志, 2011, 28(1):2407-2410.QIN X H, DUAN C J, YUAN K H, et al. A mobile health monitorsystem for the elderly[J]. Chinese Journal of Medical Physics, 2011,28(1):2407-2410.
[4]朱勇,张研,宋佳,等.基于倾角的跌倒检测方法与系统研究[J].生物医学工程学杂志, 2013, 30(1):95-99.ZHU Y, ZHANG Y, SONG J, et al. Study on the method and systemfor falling detection based on the inclinometer[J]. Journal ofBiomedical Engineering, 2013, 30(1):95-99.
[5]曹荟强,林仲志,吴水才.基于隐马尔可夫模型的老年人跌倒行为检测方法研究[J].中国生物医学工程学报, 2017, 36(2):165-171.CAO H Q, LIN Z Z, WU C S. A detection method for the fall behaviorof elders based on hidden Markov model[J]. Chinese Journal ofBiomedical Engineering, 2017, 36(2):165-171.
[6] SIE M R, LO S C. The design of a smartphone-based fall detectionsystem[C]. IEEE International Conference on Networking, Sensingand Control, 2015.
[7] ZIGEL Y, LITVAK D, GANNOT I. A method for automatic falldetection of elderly people using floor vibrations and sound-proof ofconcept on human mimicking doll falls[J]. IEEE Trans Biomed Eng,2009, 56(12):2858-2867.
[8] MERCURI M, PING J S, PANDEY G, et al. Analysis of an indoorbiomedical radar-based system for health monitoring[J]. IEEE TransMicrow Theory Tech, 2013, 61(5):2061-2068.
[9] MIRMAHBOUB B, SAMAVI S, KARIMI N, et al. Automaticmonocular system for human fall detection based on variations insilhouette area[J]. IEEE Trans Biomed Eng, 2013, 60(2):427-436.
[10]MA X, WANG H, XUE B, et al. Depth-based human fall detection viashape features and improved extreme learning machine[J]. IEEE JBiomed Health Inform, 2014, 18(6):1915-1922.
[11]FAN Y, LEVINE M D, WEN G, et al. A deep neural network for real-time detection of falling humans in naturally occurring scenes[J].Neurocomputing, 2017. DOI:10.1016/j.neucom.2017.02.082.
[12] AKAGUNDUZ E, ASLAN M, SENGUR A, et al. Silhouetteorientation volumes for efficient fall detection in depth videos[J].IEEE J Biomed Health Inform, 2017, 21(3):756-763.
[13]DANIEL H C, KANNALA J, HEIKKIL?J. Joint depth and colorcamera calibration with distortion correction[J]. IEEE Trans PatternAnal Mach Intell, 2012, 34(10):2058-2064.
[14]KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNetclassification with deep convolutional neural networks[C].International Conference on Neural Information Processing Systems,2012.
[15]LONG J, SHELHAMER E, DARRELL T. Fully convolutionalnetworks for semantic segmentation[C]. Computer Vision and PatternRecognition, 2015.
[16]KWOLEK B, KEPSKI M. Human fall detection on embedded platformusing depth maps and wireless accelerometer[J]. Comput MethodsPrograms Biomed, 2014, 117(3):489-501.