SCC: Video Based Machine Learning for Smart Traffic Analysis and Management

该项目的目标是进一步提高城市和社区部署通过更安全的交通系统拯救生命的技术的能力。方法是创建开源分析解决方案，以支持利用来自低成本视频传感器的数据的新型交通应用。使用边缘计算(执行分析而无需存储大量数据的廉价计算硬件)来处理视频数据，以减少存储的数据量。该研究项目的社会层面来自城市和大学之间的深入研究伙伴关系，目的是提供可复制的和近期的社会影响。该项目与零愿景理念相一致，以减少交通死亡，并以教育、执法和设计为基础的计划。通过自动检测险些发生的事件来了解十字路口的风险状况，社区将能够主动设计和改变街道和十字路口，使其更加安全。在解决路口、街道和系统层面的技术挑战时，设计智能城市的目标有几个研究组成部分。(I)开发新的多目标跟踪算法：将共同制定遮挡、目标特征的时间分配和目标运动的问题。(2)信号控制的综合优化与仿真：提出了信号控制参数(偏移量、相位等)的估计问题。在一个网络中作为一个整体进行优化。(Iii)当在线机器学习遇到来自伦敦金融城的真实世界反馈时，实时强化学习是一个自然的选择。我们在网络级别获取和分析连续时间数据的能力将提供有关冲突点和模式如何通过网络发生变化的洞察力。预计这将影响交通管理、智能城市规划和安全方面的决策。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The goal of this project is to further the ability of cities and communities to deploy technology that saves lives through safer transportation systems. The approach is to create open source analytics solutions to enable novel transportation applications that utilize data from low-cost video sensors. Video data are processed using edge computing (inexpensive computing hardware that performs analysis without storing significant amounts of data) in order to reduce the amount of data stored. Social dimensions of the research project emerge from the deep research partnership between the City and the University, with the goal to provide replicable and near-term social impacts. The project aligns with the Vision Zero concept to reduce traffic fatalities, with programs that are based on education, enforcement and design. By understanding the risk profile of an intersection through automated detection of near miss events, communities will be able to proactively design and alter streets and intersections to be safer. The goal of designing a smart city, when addressing the technical challenges at the intersection, street and system levels, has several research components. (i) Development of new algorithms for multi-target tracking: The problems of occlusion, temporal assignment of features to objects and target motion will be jointly formulated. (ii) Integrated optimization and simulation for signal control: We formulate the problem of estimating signal control parameters (offsets, phasing etc.) in a network as one of global optimization. (iii) Real-time reinforcement learning is a natural choice when online machine learning meets real world feedback from the City. Our ability to obtain and analyze continuous-time data at the network level will provide insights on how conflict points and patterns can change through the network. This is expected to impact decisions in traffic management, smart city planning and safety.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.