环绕摄像头环境下的视频浓缩技术研究

发布时间：2018-03-16 07:26

本文选题：视频浓缩　切入点：多摄像头　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：面对广泛应用的网络摄像头和监控摄像头,如何从海量的视频快速地找到目标成为了安防领域的重要问题。基于对象的视频浓缩技术是目前处理这一问题的主要手段。基于对象的视频浓缩技术通过提取视频中的活动物体管道,在时间轴上移动管道的位置,改变物体活动的发生时间,生成一个浓缩摘要视频,紧凑地展示物体活动内容,以达到缩短视频时间长度的目的。现有的视频浓缩技术大多是基于单摄像头监控场景,并取得了比较好的浓缩摘要结果。然而,随着多摄像头协同监控场景的涌现,现有技术不能取得令人满意的结果。为此,本论文拟研究多摄像头环绕环境下的视频浓缩技术。本论文的主要研究内容及贡献如下。1.提出一种环绕摄像头环境下的视频浓缩方法。不同于单摄像头视频浓缩方法,本方法利用多摄像头在不同角度的拍摄画面,使用概率生成地图算法和基于K最短路径优化的多目标跟踪算法,有效地跟踪物体在多摄像头监控区域内的活动,从而提取出完整的物体活动管道,进而使用能量函数改变物体活动管道的时间位置,获得浓缩的摘要结果。实验结果表明,提出的算法可以有效浓缩多摄像头监控网络中的视频。2.设计并实现了环绕摄像头环境下的视频浓缩系统。在进行摘要结果的可视化展示时,将物体以三维矩形面的形式展现在模拟的三维伪场景空间中,物体的位移活动通过移动三维矩形面在模拟地平面上的位置来体现,而物体的身体活动则通过在矩形面上渲染原视频中相应时刻的物体活动图像来表现。因此,系统用户可以改变观察的视点位置和视角方向,以减少观察视线上的物体重叠遮挡现象,获得更好的用户体验。
[Abstract]:In the face of widely used webcams and surveillance cameras, How to quickly find the target from mass video has become an important problem in the field of security. Object-based video enrichment is the main method to deal with this problem. Moving object pipes in frequency, Move the tube on the timeline, change the time when the object occurs, and generate a condensed summary video showing the object's content in a compact way. In order to shorten the length of video time, most of the existing video enrichment techniques are based on single camera surveillance scene, and get a good condensed summary result. However, with the emergence of multi-camera cooperative surveillance scene, The existing technology is not able to achieve satisfactory results. To this end, The main contents and contributions of this thesis are as follows. 1. A video concentration method under the surrounding camera environment is proposed, which is different from the single camera video concentration method. The method uses multi-camera to shoot pictures at different angles, uses probability map generation algorithm and multi-target tracking algorithm based on K shortest path optimization, effectively tracks the activity of objects in multi-camera monitoring area. In order to extract the whole moving pipe, the energy function is used to change the time position of the moving pipe, and the condensed summary results are obtained. The experimental results show that, The proposed algorithm can effectively concentrate the video in the multi-camera monitoring network. A video concentration system is designed and implemented in the surrounding camera environment. The object is displayed as a three-dimensional rectangular plane in the simulated three-dimensional pseudo-scene space, and the displacement activity of the object is reflected by moving the three-dimensional rectangular plane in the simulated ground plane. The body activity of the object is represented by rendering the image of the object moving at the corresponding time in the original video on the rectangular surface. Therefore, the user of the system can change the view position and angle of view. In order to reduce the sight of the object overlap occlusion phenomenon, to obtain a better user experience.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】