【作者】 翟艳东；

【导师】 于明；

【作者基本信息】 河北工业大学 ， 计算机应用技术， 2007， 硕士

【摘要】 步态识别是近些年来生物特征识别和计算机视觉中活跃的研究课题之一。它旨在根据人行进过程中的行走步态模式来识别其身份。它的研究主要由三部分构成:运动目标检测、特征提取和步态识别。本文针对这三部分主要开展了以下的研究工作:首先,分析了常用的运动检测算法,并根据具体情况构建背景图像,采用背景减除算法实现了运动人体的检测,利用阈值分割和形态学操作实现了图像的二值化。用Canny算法进行了边缘提取,为后续特征提取工作提供了良好的基础。其次,在特征提取方面,提出了根据步幅长度变化特征来进行周期分割的方法,在一个步态周期内确定了四个关键帧姿态。然后采用了二维离散傅里叶变换将运动检测后的二值图像变换到频域,提取四个关键帧的频谱能量幅值,计算其均值作为特征值,结合人体步幅长度特征构成五维特征向量。最后,使用标准的模式分类器——最近邻法分类器(NN)及K近邻分类器(KNN)实现身份识别。在中科院自动化所提供的CASIA步态数据库(CASIA Gait Database)进行实验,采用此种特征提取方法,在降低了算法复杂度的同时,获得了令人鼓舞的实验结果。更多还原

【Abstract】 Gait recognition is an attractive direction in biometric and computer vision in recent years. It aims to recognize individuals by the people’s walking gait pattern. A gait recognition system consists of three primary parts: moving target detection, feature extraction and gait recognition. We research on the three parts, and our work can be concluded in the following several aspects:First, we compared kinds of motion detection methods, According to our actual situation ,we chose the background subtraction methods to get the motion area. Threshold segmentation and morphologic operation were used to make the image binarization. We also used canny algorithm for edge detection, These jobs laid a good foundation for the following feature extraction works.Second, on the feature extraction aspect, we used step length feature to divide period. Two-dimensional discrete fourier transform was used to translate the binary image to the frequency domain, then we calculated the frequency energy of the four key frames, and extracted the mean value of them as the features, together with the body step length made a 5 dimensional feathers vector.At last, two supervised pattern classification technique called nearest neighbor algorithm (NN) and k-nearest neighbor algorithm (KNN) were performed for recognition. Extensive experimental results on CASIA database demonstrate that the proposed algorithm has an encouraging recognition performance with relatively lower computational cost.更多还原