【作者】 周琳琳；

【导师】 刘艳；

【作者基本信息】 天津大学 ， 计算机应用技术， 2007， 硕士

【摘要】 目前存在的各种人体运动捕获技术存在着成本高昂、环境要求较高、精确度较低等各种缺点,基于计算机视觉的人体运动捕获方法成为目前人体运动捕获研究领域的热点之一。而基于单视图视频的人体运动捕获,由于其拥有成本低、效率较高等特点,为解决一系列的基于计算机视觉的人体运动捕获问题提供了一条新的路径,能够更好地运用在小型软件的开发及应用中。本文提出一种基于计算机视觉技术的人体运动捕获方法,以简单的摄像头录制的单视图视频作为素材,以计算机图形学的各种技术作为基础,提出和改进了一些数字图像处理的算法,从图像序列中提取并描述人体轮廓的运动,然后对其进行跟踪识别,以捕获到人体运动骨架模型,最后生成人体运动骨架模型的视频动画。在算法实现过程中,首先使用差值图像分割法得到人体目标,并且提出一种孔洞区域的提取和填充算法,结合数学形态学的运算来去除人体目标的噪声影响,并强调了被忽略的人体目标图像的细节;然后使用基于梯度的彩色边缘检测方法进行人体边缘检测,使用基于数学形态学技术的边缘细化方法来进行细化边缘的提取;使用曲线曲率特征值的方法对曲线进行处理并使用直线段拟合;最后提出一种改进后的四边形拟合方法得到人体的侧影运动模型,并在此基础上得到人体运动的骨架模型。实验结果表明,本文较好地解决了基于单视频的人体运动捕获问题。进一步改善后,本文的人体运动捕获算法可以应用在智能监控、运动分析、虚拟现实技术等领域。更多还原

【Abstract】 Because of the high cost, strict condition request, and low precision of present human motion capture methods, the method based on computer vision principle becomes a hot point of this field. The human motion capture method based on single video provides a new approach to resolve the capture problem, and it adapts to the exploitations and applications in personal computer software.We present a Single Video Based Human Motion Capture method that is studied in a single video and based on some ameliorated or presented algorithms about computer graphics in this dissertation. This method obtains and describes the human motion thinning edge from image-sequence, then tracks and identifies the edge to capture the human prismatic model and creates the motion video.In this dissertation, we ameliorate and present algorithms based on computer graphic theoretic to gain human motion model. Firstly, we gain human aid area by Differential Image Segmentation, eliminate bore noise by a Bore-Area Extracting and Filling Algorithm, and eliminate edge noise by mathematical morphology. Secondly, we detect the chromatic human motion edge by a method based on grads, and thin the edge by mathematical morphology. Thirdly, we dispose curves of edge by the characters of them, and manage the curves of linear fitting. Lastly, we present a quadrangular fitting method to gain the human profile model , and on the base of that we gain human motion prismatic model.The results of our experimentation show that, we have resolved the siglevideo-based human motion capture problem in some specifically conditions. The improved method could be used in many areas.更多还原