地理场景协同的多摄像机目标跟踪研究

发布时间：2018-03-26 20:35

本文选题：GIS　切入点：视频　出处：《南京师范大学》2014年博士论文

【摘要】：地理场景具有高动态性、多尺度性和不确定性等特点,研究动态目标的实时感知方法,快速智能感知地理场景中动态目标时空特征,基于此探索动态目标的行为规律,已成为当前学术界和政府管理部门亟待解决的问题。目前,视频监控系统以其高清实时、功能智能、价格低廉等优势在安全领域发挥着越来越重要的作用。然而,现有基于监控视频的动态目标跟踪局限于二维图像空间,无法感知其在真实地理场景中的时空特征,对于具有大量视频监控设备的区域,不能实现多摄像机对动态目标的协同跟踪。鉴于此,本文以监控视频与地理场景的协同分析为手段,针对地理场景下动态目标状态的感知这一科学问题,发展地理场景约束下的动态目标时空信息提取与分析方法,取得了以下研究成果： (1)提出了多平面约束下的监控视频与2D地理空间数据几何互映射模型。现有监控视频至地理空间的映射模型假定地面为单一高程,当监控区域存在多个不同的高程平面时,现有方法需重新求解互映射矩阵,过程较繁琐。本文通过对比分析摄影测量学与计算机视觉的相机模型,构建了适用于固定摄像机与PTZ摄像机的监控视频与2D地理空间数据几何互映射模型,该模型具有各参数物理意义明确、理论严密、灵活性强等特点。 (2)提出了基于深度的监控视频与3D地理空间数据几何互映射模型。从三维空间坐标到图像坐标的转换可基于针孔成像模型来实现,但从图像坐标到三维空间坐标的转换,现有方法主要通过视线与3D模型相交的方式来实现,其计算量大且过程繁琐。本文基于3D地理空间数据可视化过程中的缓存深度值构建了监控视频与3D地理空间数据的互映射模型。与传统方法相比,本模型具有过程简洁,效率更高,可实现监控视频与3D地理空间数据间实时同步的动态互映射。 (3)提出了一种监控视频与地理空间数据的半自动互映射方法。本文系统分析了传统的基于单应的几何互映射方法,并进行了不确定性分析。鉴于地理空间数据精度日益提高,本文基于结构化地理场景的约束,设定灭点相似性、特征线相似性两个指标作为匹配的依据,探讨了监控视频与2D/3D地理空间数据视图的半自动匹配方法。 (4)设计了一种面向监控视频的前景目标时空信息提取方法。基于监控视频与地理空间数据互映射模型,提出了地理场景中监控视频前景目标时空信息的提取方法,包括目标方位信息、几何信息、目标轨迹、前景图像等,基于面向对象思想建立了其对应的时空数据模型,实现了对前景目标数据的管理与GIS集成。 (5)构建了一种基于路网约束的盲区目标轨迹估计模型。监控摄像机通常布设在相对重要的位置,具有独立性、分散性等特点,完全基于监控视频无法感知监控目标的连续运动轨迹。现有多摄像机协同下的目标连续跟踪,多利用监控视频场景间具有重叠条件下开展研究,不适合大场景中目标的连续跟踪。本文以场景中动态目标的时空信息及基础地理信息为数据基础,提出了路网(及目标行为规则)约束下的动态目标连续跟踪方法。 (6)研发了地理场景中多摄像机协同的动态目标连续跟踪原型系统。基于以上研究成果,设计并开发了动态目标连续跟踪原型系统,该系统具有监控视频与地理空间数据的互映射、动态目标时空信息提取、动态目标跟踪及可视化、大场景下监控目标轨迹估算等功能。
[Abstract]:Geographic scenes with high dynamic characteristics, multi scale and uncertainty, real-time perception of dynamic target, temporal and spatial characteristics of fast dynamic target intelligence geographic scene, so as to explore the dynamic behavior based on the target, has become the current academic circles and the government management problems to be solved.
At present, the HD video monitoring system with real-time, intelligent, low price advantage plays an increasingly important role in the field of security. However, the existing dynamic target tracking based on video is limited to 2D image space, not in the real sense of its geographical scene in the temporal and spatial features, for having a large number of video surveillance equipment area and can not realize the multi camera collaborative tracking of dynamic objects. In view of this, based on the analysis of cooperative surveillance video and geographical scene as a means of this science to solve the problem of dynamic target state under geographical scene perception, dynamic spatio-temporal information extraction and analysis method for the development of geographic scene constraints, obtained the following research results:
(1) put forward the monitoring video and 2D spatial data geometric Multi Planar Constraints mutual mapping model. The mapping model of existing video surveillance to the geographic space for a single assumed ground elevation, while there are many different elevation of the monitoring area, the existing methods need to solve mutual mapping matrix, the process is trivial. Comparative analysis of photogrammetry and computer vision camera model, constructed the monitoring video and 2D spatial data geometry for fixed camera and PTZ camera mutual mapping model, the model has clear physical meaning of each parameter, strict theory, flexibility and so on.
(2) put forward the monitoring video and 3D spatial data model based on the geometric depth of mutual mapping. Conversion from 3D coordinates to image coordinates can be realized based on pinhole imaging model, but from the image coordinate to three-dimensional space coordinate transformation, the existing methods mainly through the intersection line of sight and the model of 3D to realize the calculation. The amount is large and complicated process. The buffer depth of 3D spatial data visualization process based on the value set up a mapping model of surveillance video and 3D spatial data. Compared with the traditional method, this model has simple process, high efficiency, can realize video monitoring and 3D spatial data dynamic real time synchronization mutual mapping.
(3) proposed a surveillance video and geospatial data interoperability. Semi automatic mapping method based on the analysis of the geometry of single should be mutual mapping method based on the traditional, and the uncertainty analysis. In view of the accuracy of spatial data is increasing, the structured geographic scene based on the constraint of vanishing point, set similarity, feature line two similarity index as the matching basis, discusses the monitoring video and 2D/3D geospatial data view semi automatic matching method.
(4) design a method to extract foreground objects spatial information oriented video surveillance video and geospatial data interoperability mapping model based on the proposed extraction method of monitoring video object spatial information in geographical scene, including the target azimuth information, geometric information, target trajectory, foreground image, object oriented spatio temporal data the corresponding model is established based on the realization of the management and the GIS data of foreground object integration.
(5) constructs a blind estimation model based on target trajectory constraints. Road surveillance cameras are usually deployed in the relatively important position is independent, dispersion characteristics, based on continuous motion video monitoring target. Unable to perceive the continuous tracking of the existing multi camera collaborative target under multi use surveillance video scene with overlapping conditions to carry out research, continuous tracking is not suitable for large scene. In this paper the dynamic target in the scene and temporal information of basic geographic information data base, put forward the network (and rules) for dynamic target tracking under the constraint method.
(6) developed a prototype system of continuous tracking dynamic target geographic scene in multi camera collaboration. Based on the above research, design and development of dynamic object continuous tracking prototype system, the system of mutual mapping with monitoring video and geospatial data extraction, dynamic target spatio-temporal information, dynamic target tracking and visualization, target trajectory estimation the function of monitoring the large scene.

【学位授予单位】：南京师范大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：P208

【参考文献】