【作者】 韦素媛；

【导师】 高有行；

【作者基本信息】 西安电子科技大学 ， 计算机科学与技术， 2013， 博士

【摘要】 步态识别是进行远距离非侵犯性视觉监控、进行身份识别的理想技术之一。步态特征的提取质量对识别效果来说至关重要,在实用步态识别系统中如何构建步态数据库也是目前没有得到很好解决的问题。围绕步态识别技术与实用之间存在的这两个突出的问题,本文进行了以下几个方面的研究：(1)提出了一种建立实用步态数据库的方法,详细设计了步态特征数据库的建立方法、数据结构、步态信息的获取方法以及身高查询条件的确定方法,建立了百人规模的多信息步态数据库。(2)提出了一种结构化步态特征表示方法,结合关系数据库技术设计了一个基于数据库技术的步态识别方法,使步态识别能够利用关系数据库查询的高效性来快速缩小识别范围,提高运行效率和准确性。并通过实验验证了基于数据库技术的步态识别方法良好的身份识别效果。(3)提出一种基于时空能量图的步态特征弱周期性表示方法,给出了步态特征弱周期性和步态序列的时空能量图的定义,以统计分析的方式证明了这种特征表示对于噪声的干扰不敏感。将这种方法与论文提出的几种步态特征提取算法相结合,实验结果表明,这种方法大大简化了步态识别的预处理过程,减少步态特征的数据存储空间,降低了特征提取对于步速、摄像机采样频率等因素具体条件的依赖性,同时,识别性能也具有一定的实用参考价值。(4)利用小波多分辨率分析表征信号局部特征的优势,结合互信息熵的判据,提出一种基于小波分析和互信息熵的步态特征表示和识别算法,将小波变换得到的相似子图和各个具有局部方向特性的细节子图都利用到组合步态特征中,求得最大互信息熵时得到对应的组合特征参数组,这些参数突出了步态中行走习惯的细节差异,使提取的步态特征更集中在有利于分类的关键信息点上,从而步态识别过程更为接近人类视觉的智能识别过程。在步态识别阶段提出了互信息度量的改进方法,即沿用了能量分析和互信息度量的有效思路,又提高了识别速度。更多还原

【Abstract】 Gait recognition is one of the ideal technologies for unobtrusive security surveillance and human identification at a distance. The quantity of the extracted gait features is important to recognition. And it has not been figured out how to construct the gait database for a practical gait recognition system. Focusing on these two prominent questions between the gait recognition technology and the practical system, this dissertation mainly includes the following issues:①A method of constructing a practicality gait Database is presented. The Database constructed procedure, data structure and the obtained method of the gait are discussed in detail. Thanks for young students’passional participation, A gait database with multifarious gait information of110individuals is established. It is used for all studies in this paper.②A gait recognition method based on the database technology is proposed. The method of constructing the gait database combining with the relational database technology is presented. And the method that used the high efficien-cy of query in the relational database to reduce the individual set for the gait recognition algorithm to identify is presented. The particular design of the gait feature database, including the construction method and the data structure, the gait sampling method and the determine of the query condition about the in-dividual height range are depicted. Experimental results demonstrate that this practical gait database could availably support the gait recognition algorithm to identify the individuals more quickly and more exactly.③A gait feature expression with the semi-periodicity based on the Spatio-Temporal Energy image is described, in which the semi-periodicity and the Spatio-Temporal Energy image are defined formally. And it has been proved insensitive to noise by the statistical analysis. Experiments combining this expression with other gait feature extraction method show that this gait expression simplified the pretreatment of the gait recognition greatly, depressed the storage of the gait features, and reduced the dependence of the gait feature extraction on the step speed and the video sampling frequency, while the recognition performance possessed some practical merit.④Making use of the advantage of wavelets analysis which could express signal local feature well, and combined the mutual information merit, a bionic gait recognition based the wavelets analysis and the mutual information is proposed. The similar subimage and three detail subimages containing local orientation characters from wavelets transformation are all utilized into the combined gait feature. When the maximum mutual information gained, the optimal coefficients for the extracted combined gait feature are determined. these coefficients give prominence to particular walking habit differences of a individual, which make the combined gait feature focus on the key information discriminative for human identification and the procedure of gait recognition by computer system is more closed to the intelligent human vision.更多还原