【作者】 朱音；

【导师】 黄贤武；

【作者基本信息】 苏州大学 ， 通信与信息系统， 2003， 硕士

【摘要】 在图像处理领域，视频图像序列中运动目标的分割与跟踪是一个被广泛研究的热点。在众多的分割和跟踪方法中，本文将研究对象定位在具有一定复杂背景下单个平移运动的刚体，结合基于块的运动检测和主动轮廓模型分割目标的优点，提出了一个简单而有效的复杂背景下运动目标分割与跟踪系统，并取得了良好的效果。 本方案的实现过程为： 分割部分：对运动目标的前后三帧进行两次块匹配运动检测，通过找出两个匹配结果中运动图像块的公共部分，获得组成运动目标的图像块；求出运动目标图像块的形心作为运动目标的中心，以此中心为端点，向四周发散出角度间隔为θ的n条射线，射线的另一端终止于图像的边缘；在每条射线上按照一定的准则(点的梯度和与目标中心的距离在一定阈值范围内)找出n个初始轮廓点；以这些初始轮廓点作为主动轮廓模型(snake)的初始点，用改进的贪婪算法使snake收敛到待分割的运动目标轮廓上。 跟踪部分：根据刚体运动的特点(运动区域具有一致性和不发生较大的形变)求出运动目标的平均位移矢量，与在分割部分求出的目标的轮廓点的位置相加，作为下一帧跟踪的初始轮廓点，结合改进的snake能量函数使其更精确地收敛到运动目标轮廓上，达到准确跟踪的效果。 实验证明，本方案对于帧间位移较为明显的单个平移运动刚体，特别是外部轮廓较为平滑的物体具有良好的分割与跟踪效果。更多还原

【Abstract】 In the field of image processing, the segmentation and tracking of moving object in video sequences is a hot research topic in recent years. Various of segmentation methods can be used according to different situations of moving objects and its background.In this paper,we propose a new simple and effective system ofsegmentation and tracking of moving object in a stationary complexbackground. In this system,we combine the merits of motion detectionbased on blockmatching algorithm and the object segmentation based on active contour(snake), and get a satisfied result through the testing of a serious of different video pictures.Following are detailed steps of this system:In the segmentation section,we use three successive frames to detect the motion infomation of the pictures,by using blockmatching algorithm in two successive frames twice,we can get two pictures composed of motion blocks of the moving object,then we find out the common motion blocks of the two pictures,experiments prove these blocks approximately compose the moving object.After getting the motion blocks,we calculate the center of these blocks as the center of the moving object,thus we can use the center point as the origin,construct n straight linesjoining the boundry of the image and the center.On each line,we can find out a proper point near the object contour according to some criteria, then we use these n points as the initial points of the snake and let the snake converge on the object contour,thus finish the segmentation of the moving object.In tracking section,we still adopt the snake to converge on the object contour by improving the energy functions,in the tracking frame,initial points are acquired by calculating the average inter-frame motion vector of the block,because all snaxels have approximately thesame motion vector as the moving object is proposed to be rigid and have no salient deformation.Experiments prove that the system can be used to segment and track the single moving rigid object with comparatively large inter-frame vector,and the result is good enough especially to those objects with smooth contours.更多还原