基于HMM的动作识别结果可信度计算方法

doi:10.11959/j.issn.1000-436x.2016102

摘要/Abstract

摘要：

针对当前动作识别可信度计算方法中混淆率高、不适用于迁移学习等问题，提出一种基于样本上下文信息的可信度计算方法（S-HMM,sliding windows hidden Markov model）。该方法使用隐马尔可夫模型（HMM,hidden Markov model）理论对识别结果序列建模，将样本所在序列识别正确的概率作为识别结果的可信度，避免了当前可信度计算方法依赖于样本在特征空间中分布的问题。实验使用真实场景中的数据进行仿真，结果表明，与现有方法相比，该方法可将可信度混淆率降低37%左右。

关键词: 动作识别, 隐马尔可夫模型, 混淆率, 可信度

Abstract:

A context-based method to identify the confidence level of activ recognition was proposed,referred to as S-HMM(sliding window hidden Markov model),which reduced the confusion rate and facilitated the transfer learning.With S-HMM,the activity recognition sequence was modeled as HMM(hidden Markov model)and the corresponding probability was adopted as the confidence level.This ,S-HMM removed the dependency of the confidence level on the sample distribution in the feature space.S-HMM is extensively evaluated based on real-life activity data,demonstrat-ing a reduced confusion rate of 37% when compared to the state-of-the-art methods.

Key words: activity recognition, hidden Markov model, confusion rate, confidence level

王昌海,张建忠,徐敬东,许昱玮. 基于HMM的动作识别结果可信度计算方法[J]. 通信学报, 2016, 37(5): 143-151.

Chang-hai WANG,Jian-zhong ZHANG,Jing-dong XU,Yu-wei XU. Identifying the confidence level of activity recognition via HMM[J]. Journal on Communications, 2016, 37(5): 143-151.

图/表 12

图1

图2

图3

图4

图5

表1

表2

表3

表4

表5

图6

图7

参考文献 22

[1]	HOSEINI S A , GLUHAKA , TAFAZOLLI R . A survey on smart-phone-based systems for opportunistic user context recognition[J]. ACM Computing Surveys(CSUR), 2013,45(3): 1-27.
[2]	BULLING A , BLANKE U , SCHIELE B . A tutorial on human activity recognition using body-worn inertial sensors[J]. ACM Computing Surveys(CSUR), 2014,46(3): 57-76.
[3]	CHIANG J H , YANG P C , TU H . Pattern analysis in daily physica activity data for personal health management[J]. Pervasive and Mobile Computing, 2014,13(4): 13-25.
[4]	ZHOU B , LI Q , MAO Q , et al. Activity sequence-based indoor pede-strian localization using smartphones[J]. IEEE Transactions on Hu-man-Machine Systems, 2015,45(5): 562-574.
[5]	MARTíN H , BERNARDOS A M , IGLESIAS J , et al. Activity log-ging using lightweight classification techniques in mobile devices[J]. Personal and Ubiquitous Computing, 2013,17(4): 675-695.
[6]	SUAREZ I , JAHN A , ANDERSON C , et al. Improved activity recog-nition by using enriched acceleration data[C]// The 2015 ACM Interna-tional Joint Conference on Pervasive and Ubiquitous Computing. ACM, C2015: 1011-1015.
[7]	REISS A , HENDEBY G , STRICKER D . A novel confidence-based multiclass boosting algorithm for mobile physical acti ity monitor-ing[J]. Personal and Ubiquitous Computing, 2015,19(1): 105-121.
[8]	YOUNES R , MARTIN T L , JONES M . Activity classification at a higher level:what to do after the classifier does its best[C]// The 2015 ACM Inter-national Symposium on Wearable Computers. ACM, C2015: 83-86.
[9]	CHEN Y Q , ZHAO Z T , WANG S Q , et al. Extreme learning ma-chine-based device displacement free activity recognition model[J]. Soft Computing, 2012,16(9): 1617-1625.
[10]	DENG W Y , ZHENG Q H , WANG Z M . Cross-person activity recog-nition using reduced kernel extreme learning machine[J]. Neural Networks, 2014,53(1): 1-7.
[14]	赵中堂 . 自适应行为识别中的迁移学习方法研究[D]．北京：中国科学院大学, 2013. ZHAO Z T . Transfer learning methods for adaptive activity recogni-tion[D]. Beijing:University of Chinese Academy of Sciences, 2013.
[11]	QUINLAN J R . Induction of decision trees[J]. Machine Learning, 1986,1(1): 81-106.
[12]	COVER T M , HART P E . Nearest neighbor pattern classification[J]. IEEE Transactions on Information Theory, 1967,13(1): 21-27.
[13]	CORTES C , VAPNIK V . Support-vector networks[J]. Machine Learn-ing, 1995,20(3): 273-297.
[15]	FREDERICK J . Statistical methods for speech recognition[M]. The United States: MIT Press, 1997.
[16]	杨晓峰, 孙明明, 胡雪蕾 , 等. 基于改进隐马尔可夫模型的网络攻击检测方法[J]．通信学报, 2010,31(3): 95-101. YANG X F , SUN M M , HU X L , et al. Improved HMM model based method for detecting cyber attacks[J]. Journal on Comm nications, 2010,31(3): 95-101.
[17]	黄飞跃, 徐光祐 . 视角无关的动作识别[J]．软件学报, 2008,19(7): 1623-1634. HUANG F Y , XU G Y . Viewpoint independent action recognition[J]. Journal of Software, 2008,19(7): 1623-1634.
[18]	AGGARWAL J K , RYOO M S . Human activity analysis:a rev w[J]. ACM Computing Surveys(CSUR), 2011,43(3): 1-43.
[19]	KASTEREN T , NOULAS A , ENGLEBIENNE G , et al. Accurate activity recognition in a home setting[C]// The 10th International Con-ference on Ubiquitous Computing. Seoul, C2008: 1-9.
[20]	SHEHROZ S K , MICHELLE E K , JESSE H , et al. owards the detec-tion of unusual temporal events during activities using HMMs[C]// The 14th International Conference on Ubiquitous Computing. Seattle, C2012: 1075-1084.
[21]	FORKAN A R M , KHALIL I , TARI Z , et al. A context-aware ap-proach for long-term behavioural change detection and abnormality prediction in ambient assisted living[J]. Pattern Reco nition, 2015,48(3): 628-641.
[22]	ZHONG M , WEN J , HU P , et al. Advancing Android activit gni-tion service with Markov smoother[C]// 2015 IEEE International Con-ference on Pervasive Computing and Communication Workshops(PerCom Workshops). IEEE, C2015: 38-43.

位置	跑步	上楼	下楼	静止	走路
位置1	94	87	86	87	96
位置4	82	79	81	78	85
位置5	88	88	83	71	78
位置6	97	86	82	88	77
总计	361	340	332	324	336

动作	跑步	上楼	下楼	静止	走路
跑步	52	0	0	0	3
上楼	0	54	0	1	3
下楼	0	0	44	0	6
静止	0	1	1	71	3
走路	3	3	5	5	101
总计	55	58	50	77	116

动作	跑步	上楼	下楼	静止	走路
跑步	0.947	0.023	0.021	0.000	0.010
上楼	0.010	0.800	0.101	0.000	0.089
下楼	0.040	0.110	0.835	0.000	0.016
静止	0.000	0.000	0.001	0.990	0.009
走路	0.040	0.014	0.009	0.000	0.938

位置	ELM		SVM		BP
位置	W-ELM	S-HMM	W-SVM	S-HMM	W-BP	S-HMM
位置1	0.146	0.071	0.148	0.018	0.136	0.053
位置4	0.122	0.070	0.153	0.125	0.180	0.110
位置5	0.183	0.113	0.208	0.163	0.124	0.088
位置6	0.153	0.131	0.160	0.139	0.157	0.117
平均	0.151	0.096	0.167	0.111	0.149	0.092

位置	ELM		SVM		BP
位置	W-ELM	S-HMM	W-SVM	S-HMM	W-BP	S-HMM
位置2	0.208	0.118	0.126	0.124	0.177	0.100
位置3	0.191	0.046	0.160	0.092	0.221	0.111
位置7	0.300	0.258	0.363	0.269	0.329	0.146
位置8	0.335	0.227	0.282	0.159	0.361	0.258
平均	0.259	0.163	0.233	0.161	0.272	0.154