随着可定位智能设备的广泛使用，大量移动对象的轨迹被采集并持久化存储。这些轨迹数据已经积累到相当规模并蕴含着丰富的知识，使得以数据驱动的方式来感知并理解交通成为可能。本项目的研究目标是构建一个轨迹数据驱动的路况感知与预测方法体系，结合数据库和机器学习等手段实现对轨迹数据的有效管理，从中获取实时路况信息并挖掘交通的动态演化模式，做到对交通路况的“了如指掌”和“彰往察来”。我们将以原始轨迹数据管理、路况信息抽取、演化模式的学习与利用为主线，重点突破以下关键技术：（1）数据质量受控的轨迹大数据低负载存储机制；（2）基于实时轨迹流的即时路况在线计算方法；（3）基于反馈式多尺度深度模型的道路级路况预测；（4）预判式的交通敏感导航与信息发布策略。相关研究成果有助于构建智能化的交通控制与规划、缓解城市交通压力和提升应急管理能力，具有重要的社会及经济意义。

With the widespread use of positioning devices, the trajectories of massive moving objects are sampled and stored. These trajectory data contains valuable knowledge about traffic, and has been accumulated to a certain volume, such that it becomes practical to percept and understand traffic in a data driven approach. The aim of this project is to build systematic solutions for trajectory data driven traffic perception and prediction, by combining database and machine learning approaches to manage the data first, and then explore information and rules with respect to how traffic evolves, so that we can exactly know the current and future. Targeting to support the whole lifecycle including trajectory data management, traffic information extraction, evolution model learning and usage, we aim to investigate the following problems: (1) effective trajectory big data storage by a compressed data representation that balances redundancy and data quality issues; (2) continuous querying for real time traffic condition on top of streaming trajectories; (3) multi-scale deep learning model based traffic prediction in road granularity; (4) predictive information based traffic aware navigation and pub/sub. The outcomes of this project have valuable social and economic significance, such as intelligent traffic control, urban planning and emergency management applications.