【作者】 石欣；

【导师】 熊庆宇；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2010， 博士

【摘要】 信息网络时代,生物特征识别技术作为新兴的前沿科学技术得到很快的发展。人们逐渐在对自己的认识方式和行为方式所产生的变化中,开始了自己新的生活和新的实践。基于信息网络,人体固有的生理特性(如指纹、脸像、虹膜等)和行为特征(如笔迹、声音、步态等),通过计算系统与光学、声学、生物传感器以及生物统计学原理等科学技术,在进行个人身份鉴别和运动人体行为识别中,迅速地推进了生物特征识别理论研究和技术的发展。作为生物特征识别技术之一,步态识别旨在从运动人体的行为中,寻找和提取个体之间的变化特征,以实现自动行为识别。步态识别技术对系统分辨率要求不高,信息采集设备适应人的生活习性,具有非侵犯性以及难以隐藏等特点。国内外大量学者十分关注,积极开展深入的理论和技术研究,正逐步在智能监控、临床医学、康复治疗、运动分析、智能人工腿设计、身份识别等众多领域得到很好的应用,成为第二代生物特征识别技术的代表。本论文面向自动化领域人体行为识别的前沿研究方向,以运动人体的步态为研究对象,针对人体复杂多变运动行为的特点,基于模式识别、统计学习理论、信号处理等基础理论,从步态的采集、表征与识别三个方面开展运动人体行为识别研究。在建立步态识别系统的基础之上,重点研究了基于生理参数特征提取的步态识别算法、基于足底压力感知的跌倒行为识别算法以及基于支持向量机的运动人体行为识别算法,通过步态识别系统,进行了大量的实验测试,验证算法的有效性。主要研究成果包括以下四个方面:①系统深入地研究了国内外学者在步态识别技术和运动人体行为识别技术,对取得的理论与实践成果进行了综述研究。②研究了步态识别软硬件实验系统。基于不改变人的重力平衡和行走习惯、适应不同人的需要、便于穿戴,灵活性好的需求,研究开发了具有实时采集数据与信息传输等功能的步态识别软硬件实验系统。本系统由鞋垫压力感知模块、微处理器数据采集模块、步态数据传输模块以及后台步态数据处理模块组成。采用鞋垫式足底压力测量方法,可连续测量足底压力、时间等参数,实时监测和反馈信息,为算法研究提供了实验验证平台。③提出了基于生理参数特征提取的步态识别算法。针对足部压力测量系统输入原始数据量大、维度高、难以应用的问题,采用压力传感器获取运动人体足底压力信息,基于步态特征研究与步态特征定义,分析足底压力信息中的潜在规律,提取运动人体步态生理参数特征,建立生理参数与步态特征之间的关系模型,实现运动人体步态的自动识别。④提出了基于足底压力感知的跌倒行为识别算法。面向跌倒行为识别的实际需求,基于足底压力感知理论与识别技术,深入研究了跌倒行为压力变化特征,定义了跌倒的步态特征,抽取跌倒行为的步态特征矩阵,构建基于步态特征矩阵的二分类支持向量机模型,实现跌倒行为自动识别。并通过实验系统验证了算法的可靠性和准确性。⑤提出了基于支持向量机的运动人体行为识别算法。从人体日常运动过程的原始压力数据中分析步态特征,通过定义运动行为特征,构建了特征向量矩阵,描述并区别各种运动的压力变化特征。基于多分类支持向量机和遗传算法,构建基于遗传算法的支持向量机优化模型,实现运动人体走、跑、跳等行为的自动识别。步态识别实验系统实验结果表明,本算法识别运动人体行为具有较高的可靠性和准确性。综上所述,本文对压力感知步态、特征提取,系统构建的方法和理论研究,所提出基于生理参数特征提取的步态识别算法、足底压力感知的跌倒行为识别算法以及基于支持向量机的运动人体行为识别算法,是运动人体行为识别新的探索。步态感知启发了我们新的思考,当人们穿上特殊的鞋子,漫步于树叶、草丛之上,能够完成自己童年时的梦想。那么,信息网络将促进生物特征识别前沿技术的快速发展,新兴的多模态发展趋势,将使我们在压力感知步态识别运动人体行为研究方向上开展更加深入的研究。更多还原

【Abstract】 In information network era, biological feature recognition technology is rapidly developed as an emerging advancing front technology. People begin their new life and practice when they gradually realize the chang of cognition and behavior way. On the basis of information network, human inherent physiological property(fingerprint, face image, ins, etc.) and behavior characteristics(handwriting, voice, gait, etc.) advance the development of biological feature recognition theoretic study and technology according to technologies such as computing system and optics, acoustics, biosensor and biostatistics, in the recognition of person and human motion behavior.As one of the biological feature recognition technologies, gait recognition aims to find and extract variation between individuals from human motion behaviors to achieve automatic behavior identification. Because gait recognition technology has the characteristics of low requirement on the system resolution, data collection equipment satisfying people’s habits, non-invasive and difficult to hide, a large number of scholars aborad and at home have been deeply attracted to study on it. Gait recognition technology is gradually playing its role in intelligent monitoring, clinical, rehabilitation, movement analysis, design of intelligent artificial legs, and many other areas of identity, it has become the second-generation biometric identification technology representatives.This paper study on human motion recognition from three aspects: the collection, token and recognition of gait based on the fundamental theory of pattern recognition, statistic learning theory, signal processing, etc. On the basis of establishing the gait recognition system, it focus on the research of gait recognition algorithm based on physiological parameter feature extraction, fall recognition algorithm based on plantar pressure sensing and human motion recognition algorithm based on support vector machine. A mase of experiments have been done on gait recognition system to validate the validity of the algorithms.The main research findings include the following four areas:①The gait recognition system has been studied. The hardware of the system should not change or influence an individual’s weight balance and walking habits. Also, it should be flexible to meet different individuals’needs, be convenient to wear and be capable of data gathering and transmitting in real time. This paper researchs on the insole-based plantar pressure measuring method. The sensors is located on the positions that has to be measured. Because of the closely touch between insole and the feet, the parameters such as plantar pressure and time can be determined continuously and monitored and fed back in real time. The size of insole is adjustable,thus can be used in various kinds of shoes. The design of the insole pressure sensing module, the microcomputer-based data gathering module, the gait data transmission module and the background data processing module are also studied. At the meanwhile, the software and hardware is developed to establish a prototype system.②Gait is the motion gesture of human body during the motion process. The way to gain the information from the physiological parameters in motion to recognize the gait is the key of the study. The method to extract physiological parameter feature can solve the mass data and high dimensions problems of the input of plantar pressure measurement. This paper adopts pressure sensor to apperceive human motion, and gains human plantar pressure by contact data gathering. After the experiments and algorithm analysis, the eigenvalue of physiological parameter correlative to gait is defined. Then, the gait recognition algorithm based on physiological parameter feature extraction is proposed.③Fall is a typical gesture of human actions. The fall of elders is a dangerous action. This paper describles all the pressure change characteristics related to fall and studies on the gait feature matrix by the pressure sensing of fall events. The plantar pressure related gait feature during falling is defined and extracted. Binary support vector machine model is established and the fall recognition algorithm based on plantar pressure sensing is proposed. The experiments proved that it has high reliability and veracity on fall detection and recognition.④This paper studies on a new recognition method to apperceive human motions such as walking, running and jumpping by plantar pressure. It extracts gait feature from the initial pressure data of those motion, defines motion feature, establishs eigenvector matrix, describes and distinguishs the pressure change characteristics of motions. The optimization problem of SVM model is analyzed based on multi-classification SVM and genetic algorithm. The human motion recognition algorithm based on support vector machine is proposed. According to the validating of the experiment system, the reliability and veracity on motion recognition is proved. This method can be used to measure the amount of exercise.To sum up, the algorithms proposed in this paper are a new exploration in human motion recognition fields. Information network will advance the development of the biological feature recognition technology. Emerging multi-mode developing trend makes us a further reseach on human motion recognition by pressure sensing gait.更多还原