电视直播虚拟广告系统关键技术的研究

发布时间：2018-05-13 15:12

本文选题：虚拟广告 + 体育场地　；参考：《杭州电子科技大学》2012年硕士论文

【摘要】：随着电子产业的发展,硬件成本下降以及图像视频处理各种算法不断推陈出新,视频处理技术在工业、商业以及民用方面已经达到实用的阶段。在视频处理技术快速发展的环境下,虚拟广告系统应运而生。虚拟广告系统可广泛应用于各种体育或文艺节目的直播中,将演播室制作的虚拟广告牌插入到赛场或表演场的空地上,或用虚拟广告牌替换掉场地上原有的广告牌,合成后可达到以假乱真的效果,观众丝毫不会觉察。本文主要对体育视频中体育场地的自适应提取以及摄像机运动跟踪技术进行了研究,旨在提高虚拟广告系统的自适应性和实用性。此外,设计了一个高效的视频采集模块。论文所做的研究工作和创新点如下：首先,设计并实现了虚拟广告系统中的视频采集模块,实时采集SDI高清视频。本文采用DirectShow来实现视频的采集功能。DirectShow的优点是开发者对不同的采集卡可以做同样的应用丌发,将应用和数据传输、硬件差异以及同步问题分丌,只需要关注软件的实现,提高丌发效率,系统具有较高的通用性,丌发者可以便捷开发出多媒体应用,在实际应用中取得了良好的效果。其次,分析并提出了体育场地的自适应提取算法,包括视频帧主区域提取、二值化参数自适应以及场地线的精确提取。视频帧主区域提取是为了去除场景中包含的大量观众区域等干扰信息,提高图像检测精度和检测速度。二值化参数自适应可以在提取场地线时二值化参数自动得到阈值,使算法适用于各种场地。场地线的提取采用更为精确的基于Hough变换与最小二乘直线拟合方法,实现对场地线的自动提取。最后,分析了摄像机运动情况下球场标志线跟踪方法,实现基于自然特征点匹配的运动跟踪算法和基于场地模型的运动跟踪算法。其中基于自然特征点的运动跟踪算法比较耗时,因此,在实际项目中,本文采用了后一类方法来进行运动跟踪。在后一类方法中实现了两个方法：基于Hough空间中的直线检测方法和基于Kalman滤波的特征点预估分析方法。对比分析了在Hough空间与在x-y空间的参数预测方法。最后采用基于Kalman滤波的特征点预估分析来跟踪标志线。本文提出的体育场地的自适应提取方法和运动跟踪方法在虚拟广告系统中实现了具体应用。实例表明这几项技术在不同体育场地中具有较好的实用效果。
[Abstract]:With the development of electronic industry, the cost of hardware has decreased and various algorithms of image and video processing have been innovated. Video processing technology has reached a practical stage in industry, commerce and civil. With the rapid development of video processing technology, virtual advertising system emerges as the times require. The virtual advertising system can be widely used in the live broadcast of various sports or cultural programs. The virtual billboards made by the studio can be inserted into the field or the open space of the performance field, or the original billboards on the venue can be replaced with the virtual billboards. After synthesis can achieve the effect of false truth, the audience is not aware of the slightest. In this paper, the adaptive extraction of sports ground and the camera motion tracking technology in sports video are studied in order to improve the adaptability and practicability of the virtual advertising system. In addition, an efficient video capture module is designed. The research work and innovation of the thesis are as follows: Firstly, the video capture module in the virtual advertising system is designed and implemented to capture SDI HD video in real time. The advantage of using DirectShow to realize video capture function. DirectShow is that the developer can do the same application development for different acquisition card, open the application and data transmission, hardware difference and synchronization problem, only need to pay attention to the realization of software. In order to improve the efficiency of development, the system has high generality, and the developer can develop multimedia application conveniently, and has achieved good results in practical application. Secondly, the adaptive extraction algorithm of sports field is analyzed and put forward, including video frame main region extraction, binarization parameter adaptation and exact site line extraction. The main region extraction of video frame is to remove the large amount of interference information such as audience area contained in the scene and to improve the accuracy and speed of image detection. The adaptive binarization parameter can get the threshold value automatically when extracting the site line, which makes the algorithm suitable for all kinds of sites. A more accurate method based on Hough transform and least square straight line fitting is used to extract the site line automatically. Finally, the tracking method of the course marker line under the camera motion is analyzed, and the motion tracking algorithm based on the natural feature point matching and the motion tracking algorithm based on the site model are realized. The motion tracking algorithm based on natural feature points is time-consuming. Therefore, in the actual project, this paper uses the latter method to track motion. Two methods are implemented in the latter method: line detection method based on Hough space and feature point prediction analysis method based on Kalman filter. The methods of parameter prediction in Hough space and x-y space are compared and analyzed. Finally, the feature point prediction analysis based on Kalman filter is used to track the mark line. The adaptive extraction method and motion tracking method proposed in this paper are applied in the virtual advertising system. Examples show that these techniques have better practical effects in different sports venues.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41

【参考文献】