地理场景协同的多摄像机目标跟踪模型

The surveillance video is a kind of real-time and high-definition data source and contains a wealth of spatial and attribute information, which plays an important role not only in security but also in GIS. At present, the integration of surveillance videos and geospatial data are focused on static, one-way and interactive operation. It is difficult to meet the needs of the dynamic, two-way, automatic mapping between the surveillance videos and 2D/3D geo-spatial data. And object tracking algorithms are mainly focused on single surveillance videos and it is difficult to form the continuous trajectory of the target under a wide range which results in the failure of the object action understanding. Therefore, the project is divided into three interrelated parts. Firstly, we define a new data mapping model of surveillance videos and geo-spatial data, which includes geometry and content information. At the same time, the automatic and semi-automatic mapping methods are analyzed and tested. Secondly, based on the mapping, we analyze the common object tracking algorithm. Then, the study is focused on the problem of continuous object tracking in a larger area. We mainly propose two models including relay object tracking model and virtual object tracking model under geo-spatial constraints. The former is suited for no blind monitoring area and the latter is suited for the blind area. Thirdly, an intelligent video surveillance platform is designed and developed. Through the system, the related models and methods are validated, and integration method with current video surveillance system and GIS is discussed. The study would provide a new interactive platform for current video surveillance, expand the expression of geographic space and enrich the theory and methods of geographic information science.

监控视频作为一种实时、高清的数据源，本身蕴含着丰富的空间和属性信息，不仅对于安防，而且对于地理信息科学都具有重要作用。现今监控视频与地理空间数据映射方面，主要停留在静态、单向、交互的层次，难以满足动态、双向、自动化的互映射需求。而目标跟踪，主要面向单一监控视频，难以形成大场景下目标的连续轨迹，造成目标行为理解困难。本项目拟通过对监控视频与地理空间数据的分析，顾及几何与内容，研究监控视频与2D/3D地理空间数据的互映射模型，并探索自动几何互映射方法；在此基础上，围绕区域目标连续跟踪这个基础问题，结合时空信息，分别从无盲区目标传递跟踪和含盲区目标连续跟踪两方面进行深入研究，构建时空约束的目标连续跟踪模型；开发区域智能视频监控平台，对相关模型与方法进行验证和模拟。研究不仅为智能视频监控领域提供了新颖、高效、科学的理论与方法支持，而且拓展了地理空间的表达方式，丰富了地理信息科学的理论与方法。

视频监控系统作为一种实时获取地理空间数据的技术手段日益受到地理信息科学的重视，与之紧密相关的智慧城市建设、灾害监测、交通管理等行业也提出了迫切需求。目前，国内外针对监控视频与GIS深度融合的研究较少，较多成果停留在简单的关联、二三维可视化层面，缺乏深层次基础理论与关键技术研究，限制了视频GIS的深入应用。本项目基于GIS视角，面向GIS学科发展需求对地理场景协同的多摄像机目标跟踪问题进行了全面系统的研究。主要内容包括：a)通过对比分析计算机视觉与摄影测量学相机模型，建立了物理意义明确，适合于城市级、动态、复杂地理场景的监控视频与2D地理空间数据的互映射模型；b)通过对相机模型、三维图形绘制技术的综合分析，建立了基于深度图的监控视频与三维地理空间数据的互映射模型；c)面向结构化场景，顾及灭点和线性特征，提出了一种监控视频与地理空间数据的自动化匹配方法，为监控视频空间化提供了便利；d)基于计算机视觉中的目标检测、目标跟踪算法，融合地理场景信息，提出了一种监控视频前景目标信息提取的方法，即设计了一种基于GIS的个体目标时空信息提取方法，可提取目标的方位、几何尺寸、轨迹、前景图像等信息；e)基于监控视频空间化和光流法，提出了一种人群可定位、可量测的方法，可提取地理场景下人群真实的运动速度、方向、加速度等信息；f)结合路网、视域、目标时空信息等，对监控区域和盲区分别进行分析，建立了一个地理场景下多摄像机目标连续跟踪模型，可恢复目标的连续轨迹；开发了相应的原型系统对以上所研究的模型、算法进行了验证和应用，效果良好。本项目把监控视频作为GIS重要的数据源，重点研究了其中最为基础和关键的视频空间化和信息提取问题，相关成果可进一步丰富地理信息科学理论、技术与方法。