基于多目标跟踪的监控视频摘要系统

发布时间：2018-04-18 08:16

本文选题：视频摘要 + 目标检测　；参考：《合肥工业大学》2016年硕士论文

【摘要】：随着经济社会的快速发展,智能化的视频监控系统的快速普及,导致海量视频数据爆炸式增长。如何高效的处理、存储视频,提高视频的利用率及浏览视频的效率,一直是计算机视觉领域的研究热门。为此,本文提出一种基于多目标跟踪的视频摘要的方法。本文的主要研究工作及创新点如下：(1)本文提出基于广义关联聚类图的分层关联多目标跟踪算法。把多目标跟踪看作图的分割问题,利用广义关联聚类图的算法,在给定一个滑动时间窗口内,对某个目标每帧出现的检测图片一次性数据融合,得到自适应的轨迹片段。对轨迹片段进一步数据融合,得到完整运动轨迹,完成多目标跟踪任务。本文的算法在复杂场景下,对遮挡处理,以及身份转换问题有很好的处理能力。公共数据集上的实验结果表明,本文的方法在多目标跟踪精度和跟踪准确度上都高于主流方法。(2)本文提出一种基于多目标跟踪的视频摘要方法。将不同时刻出现的运动目标,在同一段视频帧显示,生成摘要视频。本文的视频摘要方法满足项目对实时性的要求,以多目标跟踪的运动轨迹为基础,保留运动物体的空间信息的前提下,将运动目标的检测图片重新组合整理,回贴到背景图片中得到。通过本文方法得到的摘要视频,能够有效压缩原始视频,减少视频的存储空间,缩短视频的浏览时间,提高视频的利用率。(3)本文实现了视频摘要系统软件。本文使用C++编程语言,依靠第三方软件库OpenCV,编写出以MFC为界面的视频摘要系统。系统界面简约清晰,操作容易,可以方便生成和浏览摘要视频。
[Abstract]:With the rapid development of economy and society, the rapid popularization of intelligent video surveillance system leads to the explosive growth of massive video data.How to process and store video efficiently, improve the efficiency of video utilization and browse video has been a hot topic in the field of computer vision.Therefore, this paper presents a method of video summarization based on multi-target tracking.The main research work and innovation of this paper are as follows: 1) in this paper, a hierarchical association multi-target tracking algorithm based on generalized association clustering graph is proposed.The multi-target tracking is regarded as the segmentation problem of the graph, and the algorithm of generalized association clustering graph is used. In a given sliding time window, the detection pictures appear in each frame of a target are fused in a single time, and an adaptive trajectory segment is obtained.The track segment is further fused to obtain the complete motion trajectory, and the multi-target tracking task is completed.The algorithm in this paper can deal with occlusion and identity conversion in complex scenarios.Experimental results on common data sets show that the proposed method is superior to the mainstream method in tracking accuracy and accuracy. (2) this paper proposes a video summarization method based on multi-target tracking.The moving objects at different times are displayed in the same video frame, and the summary video is generated.The video summary method in this paper meets the real-time requirements of the project. Based on the motion track of multi-target tracking and keeping the spatial information of the moving object, the detection pictures of the moving object are recombined and arranged.Return to the background image to get it.The abstract video can compress the original video effectively, reduce the storage space of the video, shorten the browsing time of the video, and improve the utilization ratio of the video.This paper uses C programming language, relying on the third party software library OpenCV, to write a video summary system with MFC interface.System interface is simple and clear, easy to operate, can easily generate and browse summary video.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN948.6

【相似文献】