视频时间戳移除和重置方法及其教育应用的研究

发布时间：2018-05-06 18:18

本文选题：时间戳移除 + 时间戳重置　；参考：《华中师范大学》2016年博士论文

【摘要】：多相机视频融合是许多智能化教育系统中的一个重要子系统。校园大型场馆的监控,教室内全景视频的录制,教育机器人多视角的视觉感知等系统都需要多个摄像机同步采集视频并合成全景视频。视频中时间戳的移除与重置是这个全景生成过程中的一个有挑战性的研究问题。当每个视频中有一个叠加的时间戳而又没有被移除的情况下,由多个视频合成的全景视频会同时出现多个时间戳。这些时间戳影响视觉效果,甚至会遮挡住一些感兴趣的对象。为了产生一个高质量的全景视频,这些时间戳需被移除,在合成的全景视频中也需要重新叠加一个时间戳。本文研究问题是如何高效地移除单个视频中出现的时间戳,并在全景视频中高质量地重置时间戳。其主要目标是研究全景视频融合功能里面的时间戳移除与重置过程中所产生的一系列问题的解决方法和关键技术,并实现这些提出的方法在教育系统中的应用。其研究内容包括六个方面:(1)时间戳数字定位;(2)时间戳区域移除与修复;(3)多相机颜色校正;(4)时间戳数字识别;(5)全景视频生成;(6)教育机器人全景视频系统的实现。通过以上六个方面内容的研究,本论文在学术上做出了一些贡献,其主要贡献有如下三点:(1)在时间戳数字区域移除与修复问题上,传统的图像修复方法一般是利用近邻区域的像素或者纹理来填充,无法还原出真实图像。针对这个难点问题,本文提出通过主动控制PTZ相机使其周期性运动移位,使被时间戳数字遮住的区域能够在时间上邻近的帧中显露出来,然后通过快速图像平移量估计算法和相邻帧像素恢复方法来实现时间戳数字区域的移除与修复。快速图像平移量估计算法采用PTZ运动估计结合全局直方图匹配来计算移位两帧之间的平移量,该方法计算量小从而使其有较好的实时性。相邻帧像素恢复方法是根据快速图像平移量估计算法求得的结果,来定位原来被时间戳数字遮挡的区域在相邻帧中的位置,并利用图像融合技术来修复原来视频帧中的时间戳区域。采用该方法移除时间戳所得到的视频具有高度的场景一致性。(2)在多相机视频颜色校正问题上,提出一个混合的颜色校正技术,该技术将区域匹配方法与伽玛校正和线性校正方法相结合,解决了相邻相机之间重叠区域不规则情况下的颜色校正问题。采用该技术进行颜色校正,即使重叠区域不规则,目标图像的色彩也能与参考图像保持高度一致。(3)在搭建的教育机器人平台上,设计并实现了全景视频系统。该系统实现了时间戳移除与重置功能。
[Abstract]:Multi-camera video fusion is an important subsystem in many intelligent education systems. The monitoring of large campus venues, the recording of panoramic video in the classroom, the visual perception of educational robots and so on require multiple cameras to synchronize the acquisition of video and synthesis of panoramic video. The removal and reset of time stamps in video is a challenging problem in the panorama generation process. When each video has a superimposed timestamp without being removed, the panoramic video synthesized by multiple videos will have multiple timestamps at the same time. These timestamps affect visual effects and may even block objects of interest. In order to produce a high quality panoramic video, these timestamps need to be removed, and a new timestamp is also needed to be overlaid in the synthesized panoramic video. In this paper, the problem is how to remove the timestamp in a single video efficiently and reset the timestamp in the panoramic video with high quality. Its main goal is to study the solutions and key technologies of a series of problems in the process of removing and resetting timestamp in the panoramic video fusion function, and to realize the application of these proposed methods in the education system. The research contents include six aspects: 1) time stamp digital location / 2) time stamp area removal and repair (3) multi-camera color correction / 4) time stamp digital recognition / 5) panoramic video generation / 6) implementation of a panoramic video system for educational robots. Through the research of the above six aspects, this paper has made some academic contributions, the main contributions are as follows: 3 points: 1) on the removal and repair of time stamp digital area, The traditional image restoration method is usually filled with pixels or textures of adjacent regions, which can not restore the real image. Aiming at this difficult problem, this paper proposes that the PTZ camera can be moved periodically by actively controlling it, so that the region hidden by the timestamp number can be exposed in the frame adjacent to the time. Then the fast image translation estimation algorithm and the adjacent frame pixel recovery method are used to realize the removal and restoration of the timestamp digital region. Fast image translation estimation algorithm uses PTZ motion estimation combined with global histogram matching to calculate the translation between two shifting frames. The pixel restoration method of adjacent frames is based on the result of fast image translation estimation algorithm to locate the position of the previously digitally occluded region in adjacent frames. The image fusion technique is used to repair the time stamp region in the original video frame. Using this method to remove the timestamp, the video has a high scene consistency. (2) A hybrid color correction technique is proposed for multi-camera video color correction. This technique combines region matching with gamma correction and linear correction to solve the problem of color correction in the case of irregular overlapped regions between adjacent cameras. Even if the overlapping area is irregular, the color of the target image can be highly consistent with the reference image. (3) on the platform of the educational robot, a panoramic video system is designed and implemented. The system realizes the function of time stamp removal and reset.
【学位授予单位】：华中师范大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：G434

【参考文献】