《运动相机下的移动物体分割算法研究》【毕业设计论文】.docVIP

摘 要 移动相机下的移动目标提取已经成为计算机视觉领域中十分热门的一个话题， 同时，也有很多的人力物力被投入到这个议题的研究实现以及改进当中来。传统上，背景减算法是被用来做目标提取的最流行的算法，因为传统上的监视系统中的相机摄像装置都是被固定安装的，这样的话，背景就是不变的而前景物体通常是动着的。然而，随着移动电话、手持摄像装置的不断普及和发展，越来越多的视频序列是被移动着的相机所拍摄的，在这些视频当中，已有的算法显然已经不适用了。 自然而然的，对那些在移动视频中提取前景目标的算法的市场需求就显得十分迫切。本毕业设计提供了卡内基梅隆大学的舍赫教授在2009年提出的移动相机下的背景减算法的一种实现。基本上本论文的实现是基于该篇论文的，但是本毕业设计在此基础上也加入了一些别的算法和策略，并做了一些修改。实验结果证明本毕业设计的方法是行得通的。 首先，SIFT特征点被从每一帧中提取出来，并且基于此，两个匹配矩阵被建立出来，而后初始的轨迹序列被通过合并这两个匹配矩阵建立出来。接下来，马萨诸塞理工学院的桑德博士提出的微粒视频技术也被加入进来用来提高轨迹的密度。接下来就是通过随即一致性算法和神经网络技术分别建立系数模型和二值标记。另外的，框级定位结果被通过稀疏模型建立出来。 关键词：目标提取；计算机视觉；背景减算法；移动相机 ABSTRACT Object detection and segmentation has long been a hot topic in the area of Computer Vision. There has been a large amount of effort undertook in this problem. Traditionally, background subtraction was the most popular method to do this job, basically because in traditional supervise system, the cameras were always static; thus, there was an assumption in traditional algorithms in such object detection problem: the background is at static and the foreground is moving. However, as the developing of mobile phones as well as digital video capture devices, there’s a growing number of video sequences are taken by such mobile devices, which old algorithms can not handle with. Consequently, the market need of algorithms to solve the object detection problem is also growing at an amazing speed due to existed algorithm can not handle those problems in moving cameras. We propose an implementation of a novel algorithm about background subtraction and object detection in moving camera settings. Basically, our method is according to a previous work of Prof. Sheikh from Carnegie Mellon University. Also, we do some modification and add some new process in this framework. The experimental results show that our method works well. Firstly, out method use SIFT to extract feature points from each frame of given video sequence and match them frame-by-frame. Secondly, the two match matrices are combined toge