Activity Recognition Approach Using a Low-Resolution Infrared Sensor
-
摘要: 如今,世界各国人口老龄化问题日益严重,为了避免独居老人发生意外,老人日常动作监测和识别算法成为了研究热点。本文设计了一种基于低分辨红外传感器的动作识别方法,通过红外传感器采集探测区的温度分布数据,对温度分布数据进行处理,从时间、温度、形变和轨迹4个方面提取多个特征,最后通过K近邻算法对“行走”、“弯腰”、“坐下”、“站起”和“摔倒”5种动作进行分类。实验结果表明平均识别准确率可达到97%,其中摔倒动作的识别准确率为100%。Abstract: The worldwide problem of population aging is becoming increasingly critical. To avoid accidents involving the elderly living alone, the study of the daily activities of the elderly using recognition and monitoring algorithms has become a research hotspot. This paper proposes an action recognition approach using a low-resolution infrared sensor. The proposed approach uses an infrared sensor to collect temperature distribution data in the detection area, and then processes the temperature distribution data, extracting multiple features in the four dimensions of time, temperature, deformation, and trajectory. Finally, the K-nearest neighbors algorithm is used to identify the five poses of "walking, " "bending, " "sitting, ""standing, " and "falling." Experimental results demonstrate that the average accuracy can reach 97% and that the accuracy for falling is 100%.
-
表 1 不同环境温度、安装空间下的阈值对比
Table 1. Comparison of thresholds under different ambient temperatures and installation spaces
Ambient temperature Laboratory Living room 22℃ 1.66 1.65 26℃ 1.71 1.73 30℃ 1.75 1.76 表 2 欧氏距离交叉验证结果
Table 2. Euclidean distance cross-validation results
Test: Train K=3 K=5 K=7 K=9 K=11 K=13 K=15 1:1 96.67% 96.8% 96.4% 96% 95.6% 95.2% 95.2% 1:4 97.33% 97.67% 97.33% 97% 96.67% 96.33% 96% 1:9 94.67% 95.33% 95.33% 95.33% 94% 93.33% 93.33% 表 3 曼哈顿距离交叉验证结果
Table 3. Manhattan distance cross-validation results
Test: Train K=3 K=5 K=7 K=9 K=11 K=13 K=15 1:1 97.73% 97.73% 97.20% 96.53% 96.93% 96.53% 96.4% 1:4 98% 97.67% 97.67% 97.67% 96.67% 96% 96% 1:9 96.67% 96% 96% 96% 95.33% 95.33% 94.68% 表 4 5种分类算法结果对比
Table 4. Comparison of the results of five classification algorithms
KNN SVM RF DT NN Bend 96.7% 92.8% 96.6% 97.9% 97.5% Stand 98.3% 97.3% 95.4% 96.3% 90.9% Sit 91.7% 75.8% 88.7% 91.3% 83.3% Fall 100% 99.1% 98.2% 96.3% 100% Walk 98.3% 95.6% 97.4% 96.7% 95.4% Accuracy 97% 92.12% 95.26% 95.7% 93.42% -
[1] 陈泰昌, 尤帅. 新时代我国人口老龄化的发展态势和战略研究[J]. 老龄科学研究, 2019, 7(4): 11-20. doi: 10.3969/j.issn.2095-5898.2019.04.002CHEN Taichang, YOU Shuai. Research on the development trends and strategies of China's population ageing in the new Era[J]. Scientific Research on Aging, 2019, 7(4): 11-20. doi: 10.3969/j.issn.2095-5898.2019.04.002 [2] 张艳梅, 马晓霞, 赫继梅, 等. 失能老人跌倒的影响因素及长期照护服务需求[J]. 中国老年学杂志, 2019, 39(17): 4355-4357. doi: 10.3969/j.issn.1005-9202.2019.17.076ZHANG Yanmei, MA Xiaoxia, HE Jimei, et al. Influencing factors of falls and demand for long-term care services in the disabled elderly[J]. Chinese Journal of Gerontology, 2019, 39(17): 4355-4357. doi: 10.3969/j.issn.1005-9202.2019.17.076 [3] 郑潇, 彭晓东, 王嘉璇. 基于姿态时空特征的人体行为识别方法[J]. 计算机辅助设计与图形学学报, 2018, 30(9): 1615-1624. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJF201809004.htmZHENG Xiao, PENG Xiaodong, WANG Jiaxuan. Human action recognition based on pose pose spatio-temporal features[J]. Journal of Computer-Aided Design & Computer Graphics, 2018, 30(9): 1615-1624. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJF201809004.htm [4] 王玉坤, 高炜欣, 王征, 等. 基于加速度传感器的人体姿态实时识别[J]. 计算机工程与设计, 2016, 37(11): 3092-3096. https://www.cnki.com.cn/Article/CJFDTOTAL-SJSJ201611041.htmWANG Yukun, GAO Weixin, WANG Zheng, et al. Real-time human activity pattern recognition based on acceleration[J]. Computer Engineering and Design, 2016, 37(11): 3092-3096. https://www.cnki.com.cn/Article/CJFDTOTAL-SJSJ201611041.htm [5] 杜英魁, 姚俊豪, 刘鑫, 等. 基于电阻式薄膜压力传感器组的人体坐姿感知终端[J]. 传感器与微系统, 2020, 39(1): 78-81. https://www.cnki.com.cn/Article/CJFDTOTAL-CGQJ202001022.htmDU Yingkui, YAO Junhao, LIU Xin, et al. Human body sitting posture sensing terminal based on resistive thin film pressure sensor groups[J]. Transducer and Microsystem Technologies, 2020, 39(1): 78-81. https://www.cnki.com.cn/Article/CJFDTOTAL-CGQJ202001022.htm [6] 李仲华, 臧春华, 杨刚. 基于噪声嵌入的跌倒检测系统的设计[J]. 微机处理, 2017(2): 74-76. https://www.cnki.com.cn/Article/CJFDTOTAL-WCLJ201702017.htmLI Zhonghua, ZANG Chunhua, YANG Gang. Design of fall detection system based on embedded noise[J]. Microprocessors, 2017(2): 74-76. https://www.cnki.com.cn/Article/CJFDTOTAL-WCLJ201702017.htm [7] Kobiyama Yuta, ZHAO Qiangfu, Omomo Kazuk, et al. Analyzing correlation of resident activities based on infrared sensors[C]//IEEE International Conference on Awareness Science and Technologys, 2015: 1-6. [8] Mashiyama S, HONG J, Ohtsuki T. Activity Recognition Using Low Resolution Infrared Array Sensor[C]//IEEE International Conference on Communications, 2015: 495-500. [9] Akira Hayashida, Vasily Moshnyaga, Koji Hashimoto, The Use of Thermal IR Array Sensor for Indoor Fall Detection[C]//IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2017: 594-599. [10] 杨任兵, 程文播, 钱庆, 等. 红外图像中基于多特征提取的跌倒检测算法研究[J]. 红外技术, 2017, 39(12): 1131-1138. http://hwjs.nvir.cn/article/id/hwjs201712011YANG Renbin, CHEN Wenbo, QIAN Qing, et al. Fall detection algorithm based on multi feature extraction in infrared image[J]. Infrared Technology, 2017, 39(12): 1131-1138. http://hwjs.nvir.cn/article/id/hwjs201712011 [11] 王召军, 许志猛, 陈良琴. 基于红外阵列传感器的人体行为识别系统研究[J]. 红外技术, 2020, 42(3): 231-237. http://hwjs.nvir.cn/article/id/hwjs202003005WANG Zhaojun, XU Zhimeng, CHEN Liangqin. Human behavior recognition system based on infrared array sensors[J]. Infrared Technology, 2020, 42(3): 231-237. http://hwjs.nvir.cn/article/id/hwjs202003005 [12] Trofimova A A, Masciadri A, Veronese F, et al. Indoor Human Detection Based on Thermal Array Sensor Data and Adaptive Background Estimation[J]. Journal of Computer and Communications, 2017, 5(4): 16-28. doi: 10.4236/jcc.2017.54002 [13] 李亚, 王颖. 基于一维最大熵的视频图像运动背景减除[J]. 计算机与现代化, 2018(3): 44-47, 53. doi: 10.3969/j.issn.1006-2475.2018.03.008LI Ya, WANG Ying. Moving background subtraction of video image based on 1-D Maximum Entropy[J]. Computer and Modernization, 2018(3): 44-47, 53. doi: 10.3969/j.issn.1006-2475.2018.03.008 [14] 薛丽霞, 罗艳丽, 王佐成. 基于帧间差分的自适应运动目标检测方法[J]. 计算机应用研究, 2011, 28(4): 1551-1552. doi: 10.3969/j.issn.1001-3695.2011.04.098XUE Lixia, LUO Yanli, WANG Zuocheng, et al. Detection algorithm of adaptive moving objects based on frame difference method[J]. Application Research of Computers, 2011, 28(4): 1551-1552. doi: 10.3969/j.issn.1001-3695.2011.04.098 [15] 黄大卫, 胡文翔, 吴小培, 等. 改进单高斯模型的视频前景提取与破碎目标合并算法[J]. 信号处理, 2015(3): 299-307. doi: 10.3969/j.issn.1003-0530.2015.03.007HUANG Dawei, HU Wenxiang, WU Xiaopei, et al. The algorithm of video foreground extraction via improved single gauss model and merge of broken targets[J]. Journal of Signal Processing, 2015(3): 299-307. doi: 10.3969/j.issn.1003-0530.2015.03.007 [16] WEI Hanchen, Hsi PinMa. A Fall Detection System Based on Infrared Array Sensors with Tracking Capability for the Elderly at Home[C]//International Conference on E-health Networking, Application & Services (Health Com), 2015: 428-434. [17] 任家东, 刘新倩, 王倩, 等. 基于KNN离群点检测和随机森林的多层入侵检测方法[J]. 计算机研究与发展, 2019, 56(3): 566-575. https://www.cnki.com.cn/Article/CJFDTOTAL-JFYZ201903012.htmREN Jiadong, LIU Xinqian, WANG Qian, et al. An multi-Level intrusion detection method based on KNN outlier detection and random forests[J]. Journal of Computer Research and Development, 2019, 56(3): 566-575. https://www.cnki.com.cn/Article/CJFDTOTAL-JFYZ201903012.htm