基于改进隐马尔科夫模型的人体步态自适应识别
发布时间：2018-10-29 18:00
【摘要】：人体步态识别技术是指利用运动学、信号处理、模式识别等理论去分析处理传感设备获得人体运动学、动力学和生理学等步态信号的技术。它在仿人机器人、人机耦合机器人(外骨骼和假肢等)、医学诊断和康复治疗、运动分析、身份识别等领域得到了广泛的应用。步态信号具有准周期性,步态阶段之间的转换可以看做一条马尔科夫链。这种性质使得隐马尔科夫模型(Hidden Markov Model,HMM)在步态阶段识别中得到了广泛的应用。但是,隐马尔科夫模型应用在步态阶段识别中存在两个不足。一是模型基于统计特征构造驻留时间分布函数,不能很好地描述步态阶段的时间特性;二是模型参数固定,未针对具体使用场景进行自适应处理。这些不足限制了步态阶段识别的效果。本文使用大腿上的加速度信号进行步态阶段识别。通过对传统隐马尔科夫模型的改进和优化,提高了模型对步态阶段识别的准确性和对步态数据的适应性。具体工作如下:1,对采集的运动步态数据进行了预处理和特征提取。预处理主要包括去噪和平滑。特征提取主要包括步态窗口划分和按窗口提取特征。2,讨论了隐马尔科夫模型原理。分析了其应用于步态阶段识别的不足,并指明了改进方向。3,在隐马尔科夫模型中引入时间参数,用驻留时间分布函数来代替自转移概率,使其能够更好的描述运动步态阶段。4,针对隐马尔科夫模型无法适应不同的穿戴者、不同的运动状态、不同的运动环境的缺陷进行改进。使用自适应算法修正模型参数,提高步态阶段识别模型的鲁棒性。5,针对自适应过程中参考模型单一的问题进行改进,提出将行为识别与步态阶段自适应识别结合的方法,实现自适应中参考模型的多样化和自主选择。我们进行了步态阶段识别的对比实验以验证以上改进的效果。结果表明针对传统隐马尔科夫模型的改进,能够提升人体步态阶段识别的效果,同时,其对不同人、不同运动模式、不同运动环境也具备一定的自适应能力。
[Abstract]:Human gait recognition technology refers to the technology that gait signals such as kinematics, signal processing and pattern recognition are obtained by means of kinematics, signal processing, pattern recognition and so on. It has been widely used in the fields of humanoid robot, human-computer coupling robot (exoskeleton and prosthesis), medical diagnosis and rehabilitation, motion analysis, identity recognition and so on. Gait signals are quasi-periodic, and the transition between gait stages can be regarded as a Markov chain. This property makes Hidden Markov Model (Hidden Markov Model,HMM) widely used in gait recognition. However, there are two shortcomings in the application of Hidden Markov Model in gait recognition. One is that the model constructs resident time distribution function based on statistical features, which can not well describe the time characteristics of gait phase; the other is that the model parameters are fixed and the adaptive processing is not carried out for specific use scenes. These deficiencies limit the effect of gait recognition. In this paper, the acceleration signal on the thigh is used to identify the gait stage. By improving and optimizing the traditional hidden Markov model, the accuracy of the model for gait recognition and the adaptability to gait data are improved. The main work is as follows: 1. Preprocessing and feature extraction of gait data are carried out. Pretreatment mainly includes denoising and smoothing. Feature extraction mainly includes gait window partition and feature extraction by window. 2. The principle of hidden Markov model is discussed. The deficiency of its application in gait recognition is analyzed, and the improvement direction is pointed out. 3. Time parameter is introduced into Hidden Markov Model, and the resident time distribution function is used to replace the self-transfer probability. 4, the hidden Markov model can not adapt to different wearer, different motion state, different motion environment defect. 4. The hidden Markov model can not adapt to the defects of different wearer, different motion state and different motion environment. 4. The hidden Markov model can not adapt to different wearer, different motion state and different motion environment. The adaptive algorithm is used to modify the model parameters to improve the robustness of the gait stage recognition model. 5. Aiming at the single problem of reference model in the adaptive process, a new method is proposed to combine the behavior recognition with the gait stage adaptive recognition. The diversity and independent selection of reference models in adaptive system are realized. A comparative experiment of gait stage recognition was carried out to verify the improved results. The results show that the improvement of traditional hidden Markov model can improve the recognition effect of human gait, at the same time, it also has certain adaptive ability to different people, different motion modes and different moving environments.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.4;O211.62
本文编号：2298392
[Abstract]:Human gait recognition technology refers to the technology that gait signals such as kinematics, signal processing and pattern recognition are obtained by means of kinematics, signal processing, pattern recognition and so on. It has been widely used in the fields of humanoid robot, human-computer coupling robot (exoskeleton and prosthesis), medical diagnosis and rehabilitation, motion analysis, identity recognition and so on. Gait signals are quasi-periodic, and the transition between gait stages can be regarded as a Markov chain. This property makes Hidden Markov Model (Hidden Markov Model,HMM) widely used in gait recognition. However, there are two shortcomings in the application of Hidden Markov Model in gait recognition. One is that the model constructs resident time distribution function based on statistical features, which can not well describe the time characteristics of gait phase; the other is that the model parameters are fixed and the adaptive processing is not carried out for specific use scenes. These deficiencies limit the effect of gait recognition. In this paper, the acceleration signal on the thigh is used to identify the gait stage. By improving and optimizing the traditional hidden Markov model, the accuracy of the model for gait recognition and the adaptability to gait data are improved. The main work is as follows: 1. Preprocessing and feature extraction of gait data are carried out. Pretreatment mainly includes denoising and smoothing. Feature extraction mainly includes gait window partition and feature extraction by window. 2. The principle of hidden Markov model is discussed. The deficiency of its application in gait recognition is analyzed, and the improvement direction is pointed out. 3. Time parameter is introduced into Hidden Markov Model, and the resident time distribution function is used to replace the self-transfer probability. 4, the hidden Markov model can not adapt to different wearer, different motion state, different motion environment defect. 4. The hidden Markov model can not adapt to the defects of different wearer, different motion state and different motion environment. 4. The hidden Markov model can not adapt to different wearer, different motion state and different motion environment. The adaptive algorithm is used to modify the model parameters to improve the robustness of the gait stage recognition model. 5. Aiming at the single problem of reference model in the adaptive process, a new method is proposed to combine the behavior recognition with the gait stage adaptive recognition. The diversity and independent selection of reference models in adaptive system are realized. A comparative experiment of gait stage recognition was carried out to verify the improved results. The results show that the improvement of traditional hidden Markov model can improve the recognition effect of human gait, at the same time, it also has certain adaptive ability to different people, different motion modes and different moving environments.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.4;O211.62
【参考文献】
相关期刊论文 前8条
1 张向刚;唐海;付常君;石宇亮;;一种基于隐马尔科夫模型的步态识别算法[J];计算机科学;2016年07期
2 徐川龙;顾勤龙;姚明海;;一种基于三维加速度传感器的人体行为识别方法[J];计算机系统应用;2013年06期
3 朱旭东;刘志镜;;基于主题隐马尔科夫模型的人体异常行为识别[J];计算机科学;2012年03期
4 石欣;熊庆宇;雷璐宁;;一种基于压力传感器的人体运动识别方法研究[J];仪器仪表学报;2010年06期
5 张文祥;张磊;马银花;;HMM自适应法在应力变异语音识别系统的应用[J];黑龙江科技学院学报;2007年05期
6 李虎生,刘加,刘润生;语音识别说话人自适应研究现状及发展趋势[J];电子学报;2003年01期
7 李虎生,杨明杰,刘润生;汉语数码语音识别自适应算法[J];电路与系统学报;1999年02期
8 郭庆，柴海新，吴文虎;隐Markov模型中状态停留时间的模型化[J];清华大学学报(自然科学版);1999年05期
相关博士学位论文 前2条
1 薛少飞;DNN-HMM语音识别声学模型的说话人自适应[D];中国科学技术大学;2015年
2 王坚;语音识别中的说话人自适应研究[D];北京邮电大学;2007年
相关硕士学位论文 前2条
1 邢秀玉;基于加速度传感器的人体步态模式分类研究[D];中北大学;2013年
2 赵欢欢;基于隐马尔可夫模型的说话人转换研究[D];中国科学技术大学;2009年
，本文编号：2298392
本文链接：https://www.wllwen.com/kejilunwen/yysx/2298392.html
