视频增强虚拟场景中PTZ相机注册与融合方法研究

发布时间：2018-06-22 14:55

本文选题：视频增强虚拟环境 + PTZ相机　；参考：《南京师范大学》2016年硕士论文

【摘要】：PTZ(Pan-Tilt-Zoom)相机在安防中广泛应用,用于安防领域的视频监控系统一般通过多路独立的视频监视器观看,由于视频间是分散、相互独立的,监控人员往往因为信息不完整连贯而产生疏漏。视频增强虚拟环境技术(AVE)通过把二维视频嵌入三维虚拟场景,实现了虚景与实景相融合,抽象与现实相结合,有利于从宏观上把握视频与场景的空间关联,为地理场景的认知提供更为全面的时空信息,更为宏观的观察视角,更好地服务安防等领域。视频图像与虚拟三维场景融合的前提是相机在虚拟场景中的注册。通过引入人工标志物是相机注册的常用方法,但这种方法难以应用于相机动态调整的场景。大多数监控相机都是可以动态调整姿态与焦距.的PTZ相机,除非厂商软件支持,无法实时获取相机调整后的姿态和焦距等信息,从而造成视频与场景无法准确融合。本文基于计算机视觉的理论和技术,对PTZ相机视频与虚拟三维场景的融合方法进行了相关研究,主要研究内容与成果如下:(1)分析了相机的成像模型与虚拟场景中虚拟相机的成像模型的关系,并明确了相机内外参数与三维渲染引擎中虚拟相机设置参数的关系,在无需引入额外的人工标志物的情况下,通过反投影方法获得3D模型世界坐标,然后利用线性方法快速估计和非线性方法迭代优化求值相结合的方法,对视频相机的初始参数进行计算,实现相机在虚拟场景中的初始注册。(2)对PTZ相机参数的实时变化进行了分析,提出了基于计算机视觉理论的PTZ相机实时跟踪与注册方法。该方法首先通过视频帧全局特征快速检测出相机参数是否发生变化,然后再使用局部特征精确匹配对相机进行实时注册,确保PTZ相机调整后视频与虚拟场景的自动准确融合。上述方法既兼顾了融合效率与融合质量,也避免了人工调整的繁琐性和滞后性。(3)在视频相机已经注册的前提下,本文分析了虚拟相机投影视频纹理的原理,并结合ShadowMap阴影算法和GPU Shader技术,基于OpensSceneGraph三维渲染引擎实现了 PTZ相机视频与虚拟三维场景的实时融合。实验结果验证了本文方法的有效性和可靠性。本文在PTZ相机实时注册方面所做的相关研究也可以推广到移动相机视频的实时注册,有助于拓展视频增强虚拟环境的应用领域。
[Abstract]:PTZ (Pan-Tilt-Zoom) camera is widely used in security. The video surveillance system used in security field is generally viewed by multiple independent video monitors. Surveillance personnel often because the information is not complete and coherent and produce omissions. Video Enhancement Virtual Environment (AVE) integrates virtual scene and real scene by embedding 2D video into 3D virtual scene, which is helpful to grasp the spatial association between video and scene from macro view. It provides more comprehensive temporal and spatial information for the cognition of geographical scene, a more macroscopic perspective of observation, and better service for the field of security and defense. The premise of fusion of video image and virtual 3D scene is the registration of camera in virtual scene. The introduction of artificial markers is a common method for camera registration, but it is difficult to apply to the scene of camera dynamic adjustment. Most surveillance cameras are able to dynamically adjust the attitude and focal length. The PTZ camera, unless supported by the vendor software, can not obtain the camera's adjusted posture and focal length in real time, resulting in the video and scene can not be accurately integrated. Based on the theory and technology of computer vision, this paper studies the fusion method of PTZ camera video and virtual 3D scene. The main contents and results are as follows: (1) the relationship between camera imaging model and virtual camera imaging model in virtual scene is analyzed, and the relationship between camera internal and external parameters and virtual camera setting parameters in 3D rendering engine is clarified. Without introducing additional artificial markers, the 3D model world coordinates are obtained by backprojection method, and then the linear method is used to estimate the world coordinates quickly and the nonlinear method is used to iteratively optimize the evaluation. The initial parameters of the video camera are calculated to realize the initial registration of the camera in the virtual scene. (2) the real-time change of the parameters of the PTZ camera is analyzed and a method of real-time tracking and registration of the PTZ camera based on the theory of computer vision is proposed. In this method, the camera parameters are quickly detected by the global feature of the video frame, and then the real time registration of the camera is carried out by using the local feature accurate matching to ensure the automatic and accurate fusion between the video and the virtual scene after the PTZ camera is adjusted. This method not only takes into account the fusion efficiency and fusion quality, but also avoids the complexity and lag of manual adjustment. (3) on the premise that the video camera is registered, this paper analyzes the principle of projecting video texture of virtual camera. Combined with ShadowMap shadow algorithm and GPU Shader technology, the real-time fusion of PTZ camera video and virtual 3D scene is realized based on OpensSceneGraph 3D rendering engine. The experimental results show that the proposed method is effective and reliable. In this paper, the research on the real-time registration of PTZ camera can also be extended to the real-time registration of mobile camera video, which is helpful to expand the application field of video enhancement virtual environment.
【学位授予单位】：南京师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP391.9

【参考文献】