| 视频图像序列运动参数估计与动态拼接 |
| |
| 引用本文： | 汤庆阳,陆佩忠.视频图像序列运动参数估计与动态拼接[J].计算机科学,2004,31(6):189-193. |
| |
| 作者姓名： | 汤庆阳 陆佩忠 |
| |
| 作者单位： | 复旦大学计算机科学与工程系,上海,200433 |
| |
| 基金项目： | 资助项目:国家自然科学基金(10171017)，国家自然科学基金重大研究计划(90204013)，教育部全国优秀博士学位论文作者专项基金，上海市科技发展基金(01JC14056) |
| |
| 摘 要： | 本文采用多重分层叠代算法来估计全局运动参数，并提出应用于动态拼接的运动分割新方法，实现既有摄像机运动又有物体运动的视频图像序列自动拼接。我们的方法基本步骤如下：首先进行全局运动参数的初始估计，并且在分层叠代过程中进行区域分类，得到初始运动模板。接着空间分割原始图像，先根据图像的空间属性由底向上分层合并图像空间区域，再利用视频图像时间属性进一步向上合并，得到图像空间分割结果。然后结合初始运动模板和图像空间分割结果，采用区域分类新方法重新对图像空间分割结果的每个区域进行分类。然后根据分类结果逐步精确求解全局运动参数。最后进行图像合成，得到全景拼接图像。我们的方法利用了多重分层叠代的优点，并且充分考虑到视频图像空间和时间上的属性，实现了运动物体和覆盖背景的精确分割，避免了遮挡问题对全局运动参数估计精度的影响。而且在图像合成时我们解决了拼接图可能产生模糊或某些区域不连续等问题。实验结果表明我们的方法实现了动态视频图像序列高质量的全景拼接。
|
| 关 键 词： | 全局运动估计 运动分割 静态拼接 动态拼接 运动模板 视频图像序列运动 参数估计 |
The Estimation of Motion Parameters and Dynamic Mosaic Representation for Videos |
| |
| TANG Qing-Yang LU Pei-Zhong.The Estimation of Motion Parameters and Dynamic Mosaic Representation for Videos[J].Computer Science,2004,31(6):189-193. |
| |
| Authors: | TANG Qing-Yang LU Pei-Zhong |
| |
| Abstract: | In this paper,we use a double hierarchical algorithm for estimating motion parameters, and a new motion segmentation method for dynamic mosaicking is proposed for creating mosaics with moving objects,which considers a moving camera. Our method consists of five main steps. In the first step,we get a initial motion mask as the result of an initial classification in the process of global motion estimation. By the second step,a image spatial segmentation algorithm uses a hierarchical and bottom-up strategy,and region merging is performed again according to temporal information of video. In the third step,we present a new method for region classification,which combines the initial motion mask and the image spatial segmentation result. By the fourth step,the precision of global motion parameters is improved using the result of region classification. In the final step,the images are composited into a panoramic mosaic. Our method can get accurate motion segmentation and region classifications as a result of the advantage of the double hierarchical algorithm and fully using spatiotemporal information in the video. Our method can also avoid inaccuracies in occlusion areas, and addresses the problem of getting a blurred and discontinuous mosaic. Experimental results show that our method achieves the aim of the creation of high quality panoramic mosaics for videos in the presence of moving objects. |
| |
| Keywords: | Global motion estimation Motion segmentation Static mosaicking Dyna-mic mosaicking Motion mask |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
点击此处可从《计算机科学》下载全文 |
