基于滑动窗口和卷积神经网络的可穿戴人体活动识别技术

何坚; 郭泽龙; 刘乐园; 苏予涵

doi:10.11999/JEIT200942

基于滑动窗口和卷积神经网络的可穿戴人体活动识别技术

doi: 10.11999/JEIT200942

何坚^{1, 2, ,},
郭泽龙¹,
刘乐园¹,
苏予涵¹

1.
北京工业大学信息学部北京 100124
2.
北京市物联网软件与系统工程技术研究中心北京 100124

基金项目: 国家重点研发计划(2020YFB2104400)，国家自然科学基金(61602016)，北京市科技计划(D171100004017003)

详细信息

作者简介:
何坚：男，1969年生，副教授，主要研究方向为智能人机交互、普适计算和物联网

郭泽龙：男，1996年生，硕士生，研究方向为智能人机交互和模式识别

刘乐园：男，1990年生，博士生，主要研究方向为物联网、机器学习、软件工程

苏予涵：男，1997年生，硕士生，研究方向为智能人机交互和模式识别

通讯作者:
何坚　Jianhee@bjut.edu.cn

中图分类号: TN911.7; TP391
计量
- 文章访问数: 1640
- HTML全文浏览量: 941
- PDF下载量: 216
- 被引次数: 0
出版历程
- 收稿日期: 2020-11-04
- 修回日期: 2021-06-02
- 网络出版日期: 2021-08-24
- 刊出日期: 2022-01-10

Human Activity Recognition Technology Based on Sliding Window and Convolutional Neural Network

HE Jian^{1, 2
, ,},
GUO Zelong¹,
LIU Leyuan¹,
SU Yuhan¹

1.
Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
2.
Beijing Engineering Research Center for IOT Software and Systems, Beijing 100124, China

Funds: The National Key R&D Program of China (2020YFB2104400), The National Natural Science Foundation of China (61602016), Beijing Science and Technology Plan (D171100004017003)

摘要

摘要: 由于缺少统一人体活动模型和相关规范，造成已有可穿戴人体活动识别技术采用的传感器类别、数量及部署位置不尽相同，并影响其推广应用。该文在分析人体活动骨架特征基础上结合人体活动力学特征，建立基于笛卡尔坐标的人体活动模型，并规范了模型中活动传感器部署位置及活动数据的归一化方法；其次，引入滑动窗口技术建立将人体活动数据转换为RGB位图的映射方法，并设计了人体活动识别卷积神经网络(HAR-CNN)；最后，依据公开人体活动数据集Opportunity创建HAR-CNN实例并进行了实验测试。实验结果表明，HAR-CNN对周期性重复活动和离散性人体活动识别的F1值分别达到了90%和92%，同时算法具有良好的运行效率。
- 人体活动识别 /
- 特征提取 /
- 卷积神经网络 /
- 滑动窗口 /
- RGB位图
Abstract: Due to the lack of unified human activity model and related specifications, the existing wearable human activity recognition technology uses different types, numbers and deployment locations of sensors, and affects its promotion and application. Based on the analysis of human activity skeleton characteristics and human activity mechanics, a human activity model based on Cartesian coordinates is established and the normalization method of activity sensor deployment location and activity data in the model is standardized; Secondly, a sliding window technique is introduced to establish a mapping method to convert human activity data into RGB bitmap, and a Convolutional Neural Network is designed for Human Activity Recognition (HAR-CNN); Finally, a HAR-CNN instance is created and experimentally tested based on the public human activity dataset Opportunity. The experimental results show that HAR-CNN achieves the F1 values of 90% and 92% for periodic repetitive activity and discrete human activity recognition, respectively, while the algorithm has good operational efficiency.
- Human Activity Recognition (HAR) /
- Feature extraction /
- Convolutional Neural Network (CNN) /
- Sliding window /
- RGB bitmap

HTML全文

图 1 基于笛卡儿坐标的人体活动力学模型

下载: 全尺寸图片幻灯片

图 2 人体活动力学数据转位图示意

下载: 全尺寸图片幻灯片

图 3 HAR-CNN架构

下载: 全尺寸图片幻灯片

图 4 Opportunity 数据集传感器分布图

下载: 全尺寸图片幻灯片

图 5 HAR-CNN网络架构

下载: 全尺寸图片幻灯片

图 6 不同滑动窗口长度和步长的F₁值

下载: 全尺寸图片幻灯片

表 1 Opportunity数据集描述

	活动	次数
周期性	站	960
	坐	169
	走路	1711
	躺	40
非周期性	开门1	125
	关门1	122
	开门2	119
	关门2	120
	开冰箱	209
	关冰箱	213
	喝饮料	136
	清理餐桌	134
	开/关灯	129
	开洗碗机	128
	关洗碗机	124
	开抽屉1	123
	关抽屉1	125
	开抽屉2	116
	关抽屉2	112
	开抽屉3	210
	关抽屉3	216

下载: 导出CSV

表 2 不同算法F₁值对比

算法	Gesture	Gesture (NULL)	ML	ML (NULL)	运行时间(s)
Bayes Network	0.79	0.81	0.82	0.74	32.9
Random Forest	0.63	0.72	0.73	0.69	21.2
Naïve Bayes	0.54	0.66	0.75	0.74	8.1
Random Tree	0.75	0.88	0.87	0.85	7.0
DeepConvLSTM	0.86	0.91	0.93	0.89	6.6
MS-2DCNN	0.81	0.89	0.92	0.85	5.8
DRNN	0.84	0.92	0.91	0.88	7.3
HAR-CNN	0.90	0.92	0.92	0.90	3.7

下载: 导出CSV

表 3 ML活动识别的准确率和召回率(%)

活动类别	DRNN准确率	DRNN召回率	HAR-CNN 准确率	HAR-CNN 召回率
NULL	90	91	92	92
站立	92	91	96	92
走路	79	82	81	94
坐下	92	85	93	73
躺平均值	90 89	85 87	90 90	85 87
标准差	5	4	5	8

下载: 导出CSV

表 4 GR活动识别的准确率和召回率(%)

活动类别	DRNN准确率	DRNN召回率	HAR-CNN 准确率	HAR-CNN 召回率
NULL 开门1	95 87	97 92	94 92	95 87
开门2	93	93	96	93
关门1	90	92	96	80
关门2	95	92	96	91
开冰箱	85	65	90	80
关冰箱	83	86	87	74
开洗碗机	88	74	83	83
关洗碗机	80	81	85	83
开抽屉1	79	77	88	86
关抽屉1	80	77	85	75
开抽屉2	77	83	97	82
关抽屉2	79	88	83	83
开抽屉3	86	85	96	90
关抽屉3	86	84	91	85
清理餐桌	93	82	100	100
喝饮料	89	88	78	95
开/关灯平均值	94 87	69 84	91 90	95 87
标准差	6	8	6	7

下载: 导出CSV

参考文献(26)

[1]	SINGH R and SRIVASTAVA R. Some contemporary approaches for human activity recognition: A survey[C]. 2020 International Conference on Power Electronics & IoT Applications in Renewable Energy and its Control, Mathura, India, 2020: 544–548.
[2]	GOTA D I, PUSCASIU A, FANCA A, et al. Human-Computer Interaction using hand gestures[C]. 2020 IEEE International Conference on Automation, Quality and Testing, Robotics, Cluj-Napoca, Romania, 2020: 1–5.
[3]	HU Ning, SU Shen, TANG Chang, et al. Wearable-sensors based activity recognition for smart human healthcare using internet of things[C]. 2020 International Wireless Communications and Mobile Computing, Limassol, Cyprus, 2020: 1909–1915. doi: 10.1109/IWCMC48107.2020.9148197.
[4]	NATANI A, SHARMA A, PERUMA T, et al. Deep learning for multi-resident activity recognition in ambient sensing smart homes[C]. The IEEE 8th Global Conference on Consumer Electronics, Osaka, Japan, 2019: 340–341.
[5]	RODRIGUES R, BHARGAVA N, VELMURUGAN R, et al. Multi-timescale trajectory prediction for abnormal human activity detection[C]. 2020 IEEE Winter Conference on Applications of Computer Vision, Snowmass, USA, 2020: 2615–2623.
[6]	DEEP S and ZHENG Xi. Leveraging CNN and transfer learning for vision-based human activity recognition[C]. The 29th International Telecommunication Networks and Applications Conference, Auckland, New Zealand, 2019: 1–4.
[7]	邓诗卓, 王波涛, 杨传贵, 等. CNN多位置穿戴式传感器人体活动识别[J]. 软件学报, 2019, 30(3): 718–737. doi: 10.13328/j.cnki.jos.005685 DENG Shizhuo, WANG Botao, YANG Chuangui, et al. Convolutional neural networks for human activity recognition using multi-location wearable sensors[J]. Journal of Software, 2019, 30(3): 718–737. doi: 10.13328/j.cnki.jos.005685
[8]	WANG Yan, CANG Shuang, and YU Hongnian. A survey on wearable sensor modality centred human activity recognition in health care[J]. Expert Systems with Applications, 2019, 137: 167–190. doi: 10.1016/j.eswa.2019.04.057
[9]	BAO Ling and INTILLE S S. Activity recognition from user-annotated acceleration data[C]. The 2nd International Conference on Pervasive Computing, Vienna, Austria, 2004: 1–17.
[10]	NURMI P, FLORÉEN P, PRZYBILSKI M, et al. A framework for distributed activity recognition in ubiquitous systems[C]. International Conference on Artificial Intelligence, Las Vegas, USA, 2005: 650–655.
[11]	RAVI N, DANDEKAR N, MYSORE P, et al. Activity recognition from accelerometer data[C]. The Twentieth National Conference on Artificial Intelligence and the Seventeenth Innovative Applications of Artificial Intelligence Conference, Pittsburgh, USA, 2005: 1541–1546.
[12]	何坚, 周明我, 王晓懿. 基于卡尔曼滤波与k-NN算法的可穿戴跌倒检测技术研究[J]. 电子与信息学报, 2017, 39(11): 2627–2634. doi: 10.11999/JEIT170173 HE Jian, ZHOU Mingwo, and WANG Xiaoyi. Wearable method for fall detection based on Kalman filter and k-NN algorithm[J]. Journal of Electronics &Information Technology, 2017, 39(11): 2627–2634. doi: 10.11999/JEIT170173
[13]	TRAN D N and PHAN D D. Human activities recognition in android smartphone using support vector machine[C]. The 7th International Conference on Intelligent Systems, Modelling and Simulation, Bangkok, Thailand, 2016: 64–68.
[14]	HANAI Y, NISHIMURA J, and KURODA T. Haar-like filtering for human activity recognition using 3D accelerometer[C]. The 13th Digital Signal Processing Workshop and 5th IEEE Signal Processing Education Workshop. Marco Island, USA, 2009: 675–678.
[15]	ORDÓÑEZ F J and ROGGEN D. Deep convolutional and LSTM recurrent neural networks for multimodal wearable activity recognition[J]. Sensors, 2016, 16(1): 115. doi: 10.3390/s16010115
[16]	何坚, 张子浩, 王伟东. 基于ZigBee和CNN的低功耗跌倒检测技术[J]. 天津大学学报: 自然科学与工程技术版, 2019, 52(10): 1045–1054. doi: 10.11784/tdxbz201808059 HE Jian, ZHANG Zihao, and WANG Weidong. Low-power fall detection technology based on ZigBee and CNN algorithm[J]. Journal of Tianjin University:Science and Technology, 2019, 52(10): 1045–1054. doi: 10.11784/tdxbz201808059
[17]	GJORESKI H, LUSTREK M, and GAMS M. Accelerometer placement for posture recognition and fall detection[C]. 2011 Seventh International Conference on Intelligent Environments, Nottingham, UK, 2011: 47-54.
[18]	WANG Changhong, LU Wei, NARAYANAN M R, et al. Low-power fall detector using triaxial accelerometry and barometric pressure sensing[J]. IEEE Transactions on Industrial Informatics, 2016, 12(6): 2302–2311. doi: 10.1109/TII.2016.2587761
[19]	GJORESKI H, KOZINA S, GAMS M, et al. RAReFall - Real-time activity recognition and fall detection system[C]. 2014 IEEE International Conference on Pervasive Computing and Communication Workshops, Budapest, Hungary, 2014: 145-147.
[20]	SHOTTON J, FITZGIBBON A, COOK M, et al. Real-time human pose recognition in parts from single depth images[C]. The CVPR 2011, Colorado, USA, 2011: 1297–1304. doi: 10.1109/CVPR.2011.5995316.
[21]	KONIUSZ P, CHERIAN A, and PORIKLI F. Tensor representations via kernel linearization for action recognition from 3D skeletons[C]. The 14th European Conference on Computer Vision, Amsterdam, The Netherlands, 2106: 37–53.
[22]	LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278–2324. doi: 10.1109/5.726791
[23]	KRIZHEVSKY A, SUTSKEVER I, and HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84–90. doi: 10.1145/3065386
[24]	SIMONYAN K and ZISSERMAN A. Very Deep convolutional networks for large-scale image recognition[J]. arXiv: 1409.1556, 2015.
[25]	Opportunity dataset[EB/OL]. https://archive.ics.uci.edu/ml/datasets/OPPORTUNITY+Activity+Recognition, 2015.
[26]	MURAD A and PYUN J Y. Deep recurrent neural networks for human activity recognition[J]. Sensors, 2017, 17(11): 2556. doi: 10.3390/s17112556