毕业论文（设计）人体运动目标提取和跟踪.docVIP

┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ 装 ┊ ┊ ┊ ┊ ┊ 订 ┊ ┊ ┊ ┊ ┊ 线 ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ ┊ PAGE \* MERGEFORMAT II PAGE \* MERGEFORMAT II 摘 要 视频图像的特征提取与运动跟踪是计算机视觉的一个热门研究方向，而人体运动目标特征的提取和跟踪更是有着广泛的应用前景和研究价值。现在，一些研究人员已研究出许多有着实际应用价值的理论和方法。但针对人体运动的特点和受到光线变化、旋转和噪声等外界变化干扰时，如何更加准确的提取人体运动特征和跟踪运动趋势，依然是一个亟待解决的问题。本文主要针对帧差法和Mean Shift算法进行人体特征提取与运动跟踪时分别出现的问题进行研究。 帧差法对人体运动进行特征提取时存在如下问题:即两帧间当人体运动变化部分有重叠时会出现空洞问题。为此，本文提出了结合直方图图像增强技术的二帧差分法来解决上述问题。利用直方图分段线性变换技术对视频图像序列进行动态范围调整，突出每一帧图像中感兴趣的人体目标部分，再对这一部分利用二帧差分法进行提取。 Mean Shift算法在跟踪人体运动目标时，带宽的固定影响了跟踪效果，甚至导致了跟踪的失败。针对这一问题，本文对经过预处理后人体头部二值图像所呈现的圆形，由Hough变换检测出其半径的大小。再计算出半径的缩放比例，进一步由该缩放比例来调整带宽。改进后的Mean Shift算法在对人体远离或者靠近摄像头的运动跟踪中，带宽能够随着人体运动自动进行缩放变化，增强了跟踪的准确度。 本文提出了基于直方图的二帧差分法的和基于人体头部特征的Mean Shift算法，并在各章节给出了实验结果和结果分析。实验结果验证了本文算法的准确性和实时性。 关键词:帧差法，Mean Shift算法，直方图图像增强技术，Hough变换 PAGE \* MERGEFORMAT III PAGE \* MERGEFORMAT III ABSTRACT Feature extraction and motion tracking in video images are a popular computer vision research. However,the body moving target feature extraction and motion tracking have broad application prospects and research value. Now, some researchers have developed many theories and methods of the practical application . But how to more accurately extract human motion characteristics and track movement trends to the interference of external changes that such as the characteristics of human motion and by the light changes, rotation, and noise is still an unresolved issue. In this paper, my research focuses some questions that the frame difference method and the Mean Shift algorithm are used to extract and track the human motion. There will be a question when using frame difference method to extract the feature of human motion: Cavitation will appear when human motion occurs overlap between the two frames. So we propose two frames difference method combined with the histogram of image enhancement technology to solve these problems. Using the technique of histogram piecewise linear transformation to adjust the video image sequence in the