CRII: NeTS: Ubiquitous Sensing based Location-aware Driving Safety System

该项目利用移动传感和车辆定位来识别细粒度的异常驾驶行为，如摇摆、转弯和快速掉头，并进一步推断位置感知的危险车辆状态。现有的几项工作都试图通过基于预先部署的基础设施(如酒精传感器、红外传感器和摄像头)检测司机的状态来检测异常驾驶行为。这种方法会产生额外的安装费用，因此很难被广泛采用。为了构建普适的位置感知驾驶安全系统，本项目尝试采用低功耗感知(利用用户在车辆上携带的移动设备)和基于统计分析的学习技术来定位车辆并识别细粒度的异常驾驶行为。更重要的是，该系统持续跟踪驾驶员的行为，并确定细粒度的位置相关危险车辆状态，如在双向道路中心线上驾驶或长期占用左侧车道。该项目旨在进行全面研究，以了解当前移动设备在多大程度上能够模拟各种现实世界的驾驶行为和相应的车辆动力学。为了解决驾驶安全问题，开发了一种新的实时移动感知系统，该系统结合了实时移动感知和异质驾驶环境。最终结果将是解决复杂环境的动态影响并为物联网(IoT)提供明确指导的网络物理架构的持久原则。具体地说，从能够描述每种类型异常驾驶行为的移动传感器读数中调查有效的特征。因此，可以提取这些特征来定位车辆，并在考虑一般驾驶场景和异类移动设备的情况下导出异常驾驶行为(例如，摇摆、转向、快速掉头和突然刹车)的模式。开发了基于机器学习的技术来生成分类器模型，该模型可以清楚地识别细粒度的异常驾驶行为。该分类器模型将进一步作为设计位置感知驾驶安全系统的基础，该系统可以跟踪用户的驾驶行为，并使用低计算能力的移动设备实时实现与位置相关的危险车辆状态。

This project exploits mobile sensing and vehicle localization to identify fine-grained abnormal driving behaviors, such as weaving, swerving, and fast U-turns, and further to infer location-aware dangerous vehicular status. Several existing works have tried to detect abnormal driving behaviors by focusing on detecting drivers' status based on pre-deployed infrastructure, such as alcohol sensors, infrared sensors, and cameras. Such approaches incur extra installation cost and are thus difficult to be widely adopted. In order to build pervasive location-aware driving safety systems, this project tries to deploy low power consumption sensing (utilizing mobile devices carried by users in vehicles) and learning techniques based on statistical analysis to localize vehicles and identify fine-grained abnormal driving behaviors. More importantly, the proposed system keeps tracking the drivers' behaviors and determines fine-grained location-related dangerous vehicular status, such as driving on the center line of two-way roads or occupying left lanes for a long time.This project seeks to conduct a comprehensive study to understand to what extent the current mobile devices can model various real-world driving behaviors and corresponding vehicle dynamics. A new real-time mobile sensing system, which combines real-time mobile sensing and heterogeneous driving environments, is developed to address driving safety concerns. The final results will be the abiding principles of cyber-physical architecture that resolve dynamic impacts of complex environments and provide clear guidelines over Internet of Things (IoTs). Specifically, effective features are investigated from mobile sensor readings that are able to depict each type of abnormal driving behaviors. These features can thus be extracted to localize the vehicles and derive the patterns of abnormal driving behaviors (e.g., weaving, swerving, fast U-turn, and sudden breaks) with the consideration of generic driving scenarios and heterogeneous mobile devices. Techniques based on machine learning are developed to generate a classifier model that could clearly identify fine-grained abnormal driving behaviors. The classifier model will be further utilized as a foundation to devise the location-aware driving safety system, which can track users' driving behaviors and realize location-related dangerous vehicular status in real-time using low-computing-capability mobile devices.