﻿ 基于姿态估计的实时跌倒检测算法
 控制与决策  2020, Vol. 35 Issue (11): 2761-2766 0

### 引用本文 [复制中英文]

[复制中文]
YU Nai-gong, BAI De-guo. Real-time fall detection algorithm based on pose estimation[J]. Control and Decision, 2020, 35(11): 2761-2766. DOI: 10.13195/j.kzyjc.2019.0382.
[复制英文]

### 文章历史

1. 北京工业大学信息学部, 北京100124;
2. 计 算智能与智能系统北京重点 实验室, 北京100124;
3. 数字社区教育部工程研 究中心, 北京100124

Real-time fall detection algorithm based on pose estimation
YU Nai-gong , BAI De-guo
1. Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China;
2. Beijing Key Laboratory of Computational Intelligence and Intelligent System, Beijing 100124, China;
3. Digital Community Ministry of Education Engineering Research Center, Beijing 100124, China
Abstract: In order to quickly and accurately detect the occurrence of falls in the elderly, this paper presents a real-time fall detection algorithm based on pose estimation.Firstly, the human pose estimation algorithm based on deep learning is used to obtain the coordinates of the joint point.Then, by calculating the falling speed of the centroid point when the human body falls, whether the ordinate value of the neck joint point after the fall is greater than the threshold, and the relative positional relationship of the shoulder-waist joint point in the image, whether the fall occurs is determined.The algorithm uses a monocular camera to detect, which is easily used in an embedded way for robots.The experimental results show that compared with the current advanced methods, the proposed algorithm has achieved good results.
Keywords: fall detection in real-time    deep learning    human pose estimation algorithm    monocular camera    robots    embedding
0 引言

1 跌倒检测算法

 图 1 姿态估计算法获取到的18个关节点

1.1 单目相机安装在固定位置处的跌倒检测

 图 2 跌倒检测算法判定步骤
1.1.1 判定条件1

 (1)

 (2)

 (3)

1.1.2 判定条件2

vvth作为判定跌倒是不充分的, 这时会有很大的误判情况, 例如人体下蹲系鞋带、捡物品等.因此, 为了排除上述情况, 在人体下蹲时获取1关节点在图像中的最大高度Hth.判定条件1满足后每10帧获取1次y1, 如果在一定时间内y1始终大于Hth, 则说明可能发生了跌倒.判定条件2的表达式为

 (4)

 (5)

 (6)

 (7)
 (8)

 (9)
 (10)

1.1.3 判定条件3

 (11)
 图 3 判定条件3示意

 (12)

1.2 单目相机安装在机器人设备上的跌倒检测

2 实验结果与分析

2.1 相机安装在固定位置处的实验

 图 4 相机在固定位置时的检测结果

2.2 相机安装在机器人上的实验

 图 5 相机在机器人设备上的检测结果

1) 真阳性(true positive):是跌倒的实验中被检测为跌倒的次数;

2) 假阳性(false positive):不是跌倒的实验中却被检测为跌倒的次数;

3) 真阴性(true negative):不是跌倒的实验中没有被检测为跌倒的次数;

4) 假阴性(false negative):是跌倒的实验中却没有被检测为跌倒的次数.

 (13)
 (14)
 (15)

1) 改善人体姿态估计算法, 增加人体骨架图检测的精度, 进一步降低误报率;

2) 单个相机视野不开阔, 且在有障碍物的情况下有时检测不到人体骨架图, 因此可尝试用多个相机从多角度检测, 以提高跌倒检测的成功率;

3) 模拟跌倒场景种类较为简单, 需增加各种场景下的跌倒实验.

3 结论

 [1] Burns E R, Kakara R. Deaths from falls among persons aged ≥65 years - United States, 2007-2016[J]. Morbidity and Mortality Weekly Report, 2018, 67(18): 509-514. DOI:10.15585/mmwr.mm6718a1 [2] 耳玉亮, 段蕾蕾, 叶鹏鹏, 等. 2014年全国伤害监测系统老年人非故意伤害病例特征分析[J]. 中国健康教育, 2016, 32(4): 312-317. (Er Y L, Duan L L, Ye P P, et al. Analysis on the characteristics of unintentional injuries among the elderly from Chinese national injury surveillance system, 2014[J]. Chinese Journal of Health Education, 2016, 32(4): 312-317.) [3] Noury N, Rrmeau P, Bourke A K, et al. A proposal for the classification and evaluation of fall detectors[J]. IRBM, 2008, 29(6): 340-349. DOI:10.1016/j.irbm.2008.08.002 [4] Mohamad Daher, Maan El Badaoui El Najjar, Ahmad Diab, et al. Automatic fall detection system using sensing floors[J]. International Journal of Computing & Information Sciences, 2016, 12(1): 75-82. [5] Minvielle L, Atiq M, Serra R, et al. Fall detection using smart floor sensor and supervised learning[C]. The 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). Seogwipo, 2017: 3445-3448. [6] Li Y, Ho K C, Popescu M. A microphone array system for automatic fall detection[J]. IEEE Transactions on Biomedical Engineering, 2012, 59(5): 1291-1301. DOI:10.1109/TBME.2012.2186449 [7] Wang F T, Chan H, Hsu M, et al. Threshold-based fall detection using a hybrid of tri-axial accelerometer and gyroscope[J]. Physiological Measurement, 2018, 39(10): 105002. DOI:10.1088/1361-6579/aae0eb [8] Ranakoti S, Arora S, Chaudhary S, et al. Human fall detection system over IMU sensors using triaxial accelerometer[J]. Computational Intelligence: Theories, Applications and Future Directions: Volume I, 2018, 798: 495-507. [9] Kumar V S, Acharya K G, Sandeep B, et al. Wearable sensor-based human fall detection wireless system[J]. Wireless Communication Networks and Internet of Things, 2018, 493: 217-234. [10] Mehmood A, Nadeem A, Ashraf M, et al. A novel fall detection algorithm for elderly using SHIMMER wearable sensors[J]. Health and Technology, 2019, 9(4): 631-646. DOI:10.1007/s12553-019-00298-4 [11] Sara Zulj, Goran Seketa, Igor Lackovic. Accuracy comparison of ML-based fall detection algorithms using two different acceleration derived feature vectors[C]. IFMBE Proceedings of World Congress on Medical Physics and Biomedical Engineering 2018. Singapore: Springer, 2018: 481-485. [12] 宋菲, 薛质. 基于Open CV的老人跌倒检测的设计和实现[J]. 信息技术, 2015, 39(11): 137-139. (Song F, Xue Z. Design and implementation of fall detection for elder person based on Open CV[J]. Information Technology, 2015, 39(11): 137-139.) [13] Yu M, Gong L, Clifford R, et al. A monocular camera based person-specific fall detection system exploiting deep neural network aided unsupervised[C]. IEEE International Conference on Acoustics, Speech, and Signal Processing. Brighton, 2019. [14] Patsadu O, Watanapa B, Dajpratham P, et al. Fall motion detection with fall severity level estimation by mining kinect 3D data stream[J]. The International Arab Journal of Information Technology, 2018, 15(3): 378-388. [15] 王君泽, 朱小龙, 瞿畅. 基于Kinect骨架追踪的人体摔倒自动检测[J]. 上海交通大学学报, 2015, 49(9): 1359-1365. (Wang J Z, Zhu X L, Qu C. Automatic fall detection using human skeleton tracking algorithm based on kinect sensor[J]. Journal of Shanghai Jiaotong University, 2015, 49(9): 1359-1365.) [16] De Miguel K, Brunete A, Hernando M, et al. Home camera-based fall detection system for the elderly[J]. Sensors, 2017, 17(12): 2864. DOI:10.3390/s17122864 [17] Ma C, Shimada A, Uchiyama H, et al. Fall detection using optical level anonymous image sensing system[J]. Optics & Laser Technology, 2019, 110: 44-61. [18] Sumiya T, Matsubara Y, Nakano M, et al. A Mobile robot for fall detection for elderly-care[J]. Procedia Computer Science, 2015, 60: 870-880. DOI:10.1016/j.procs.2015.08.250 [19] Sun P F, Ming A, Yao C, et al. Fall detection system based on mobile robot[J]. Image and Graphics Technologies and Applications, 2018, 875: 263-271. [20] Cao Z, Simon T, Wei S E, et al. Realtime multi-person 2D pose estimation using part affinity fields[C]. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Hawaii, 2017: 7291-7299. [21] Glen Debard, Peter Karsmakers, Mieke Deschodt, et al. Camera based fall detection using multiple features validated with real life video[C]. Workshop Proceedings of the 7th International Conference on Intelligent Environments. Nottingham, 2011: 441-450. [22] Triantafylou D, Krinidis S, Ioannidis D, et al. A real-time fall detection system for maintenance activities in indoor environments[J]. IFAC-PapersOnLine, 2016, 49(28): 286-290. DOI:10.1016/j.ifacol.2016.11.049 [23] Geertsema E E, Visser G H, Viergever M A, et al. Automated remote fall detection using impact features from video and audio[J]. Journal of Biomechanics, 2019, 88: 25-32. DOI:10.1016/j.jbiomech.2019.03.007 [24] Panahi L, Ghods V. Human fall detection using machine vision techniques on RGB-D images[J]. Biomedical Signal Processing and Control, 2018, 44: 146-153. DOI:10.1016/j.bspc.2018.04.014