海量移动对象情境查询与分析预测关键问题研究

With the continuous development of spatial-temporal database technology and wireless mobile communication technology, the research and application of temporal information of spatial objects become increasingly important. Currently, the moving objects database technology has been applied to urban planning, traffic management, physical distribution, mobile e-commerce, weather forecast, etc. Facing the limitations of existing technologies caused by the characteristics of moving objects (such as dynamics, uncertainty), this project investigates into the key problems in moving objects database. The main contents are: efficient query utilizing context-aware semantic analysis, trajectory prediction based on trajectory structures and Gaussian Mixture Models (GMM), and visualization technology integrating moving-mode detection and context-awareness. Firstly, for efficient query, we will use the ontology matching and keywords recommendation algorithms to combine geospatial information for keyword disambiguation and providing a narrower scope to the search, which will improve the query efficiency. Meanwhile, privacy-preserving processing will be considered in query as well. Secondly, to achieve adaptive and robust trajectory prediction, we will extract structural features of trajectories and build prediction algorithms based on GMM. Based on these, the semi-supervised learning framework is adopted for trajectory analysis and prediction. Thirdly, for visualizing the dynamic information of the real world in the virtual world, we will use pattern matching and ontology hierarchies to efficiently visualize the information based on automatic construction of spatial-temporal dynamic context and profiles. Finally, we will realize the prototype system based on the above research, and conduct the research in the applications based on the Platform of Intelligent Vehicle Management and Service in Xiamen city. The project has high academic values and is easy to extend to related areas and various applications.

随着时空数据库技术和无线移动通信技术的不断发展，对空间对象时域信息的研究和应用显得日益重要，目前移动对象数据库技术广泛应用在城市规划、交通管理、物流配送、移动电子商务、天气预报等场景。针对移动对象的动态特性和不确定性等特点所造成的技术局限性，项目重点研究结合情境语义分析的高效查询、基于轨迹结构和高斯混合模型的预测方法以及融合移动模式检测与情境感知的可视化技术等关键问题。结合地理空间信息，通过本体匹配和关键词推荐等情境语义分析技术，消除关键词歧义，缩小范围，提高查询效率并关注隐私保护；提取轨迹结构特征，利用半监督学习技术，构建基于高斯混合模型轨迹预测算法，确保适应性和鲁棒性；基于时空动态情境和轮廓的自动构建技术，使用模式匹配和本体论层次结构，尝试解决如何在虚拟世界中高效可视化真实世界中的动态信息。最后实现原型系统并以厦门市车辆智能管理与服务平台为基础开展深入应用研究，学术价值较高，易于扩展。

本项目按照既定的研究目标，项目组充分利用与厦门卫星定位应用股份有限公司、厦门科拓通讯技术股份有限公司、厦门亿联网络技术股份有限公司等的合作，充分结合企业的数据优势与海量移动对象大数据，针对移动对象高效查询、不确定性轨迹预测、计算机视频跟踪等关键技术进行深入研究。在移动对象高效查询方法，面向基于地理密度相关性、基于多源空间数据的区域重要性信息、基于空间数据聚类信息和基于动态交通网络信息的多种环境，解决从空间搜索到跨域搜索和多样化路径查询的效率问题；在不确定性轨迹预测方面，基于移动对象的准确分类、客流量常规影响因素的特征提取、城市环境影响效应分析，深入研究交通流量预测、移动对象运行时间与站点流量预测等技术，提升不确定性轨迹预测的准确率；在计算机视频跟踪方面，结合情境感知对象跟踪技术，研究提升对象跟踪算法速度和性能的技术；通过综合运用上述技术，形成一个原型系统，并应用于面向城市交通场景下的大规模交通数据分析与服务平台，产生一定的经济价值与学术价值。相应的研究成果共发表学术论文22篇（SCI/EI检索20篇）、授权发明专利 2 项、申请发明专利 3 项、获得 2 项软件著作权，达到预期研究成果。同时，提交年度报告3份、结题报告1份。