Urban Scene Analytics for Road Safety

道路安全城市场景分析

• 批准号: 560312-2020
• 负责人: Greenspan, Michael
• 金额: $ 9.62万
• 依托单位: Queen's University
• 依托单位国家: 加拿大
• 项目类别: Alliance Grants
• 财政年份: 2021
• 资助国家: 加拿大
• 起止时间: 2021-01-01 至 2022-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=721586
• 关键词: Urban; Scene; Analytics; Road; Safety

This project will develop a capability to automatically recognize safety critical events and conditions within a stream of image and video data, acquired from a network of mobile dash cam and stationary street cam sensors, with the aim of improving road safety. The data source will be a large, distributed network of cameras, comprising both mobile dash cams deployed on Geotab-enabled vehicles, and stationary street cams such as are available from municipal infrastructure. Processing of this data will rely on state-of-the-art Machine Learning methods, primarily variations of Deep Convolutional Neural Networks. The outcome of the project will be the development of advanced deep learning based methods to recognize events and characteristics in urban scene data which are related to road safety, such as an unsafe pedestrian crossing or high-volume intersections, or degraded municipal infrastructure such as burned-out street lights or potholes. The project outcome will include a thorough characterization of the effectiveness of these methods, through the design of a set of structured experiments. Further, these methods will be implemented in a standardized development environment, to facilitate transfer to Geotab for subsequent commercialization, and to demonstrate the applicability to the City of Kingston for municipal infrastructure monitoring. The project impact will be to advance the use of machine learning methods to detect safety-critical events and scenarios within images and video of urban scenes. The project has the potential to improve the safety of urban environments, both through the development of advanced driver and/or pedestrian alerts as safety-critical events and conditions are recognized in real-time, as well as through informing the design criteria of roads and associated infrastructure as the root causes of certain events are revealed. Further tangible impact of the project will be novel product offerings from Geotab, as well as identifying opportunities for the City of Kingston to apply these techniques for infrastructure asset tracking.

该项目将开发一种自动识别从移动的仪表盘摄像头和固定街道摄像头传感器网络获取的图像和视频数据流中的安全关键事件和条件的能力，以提高道路安全。数据源将是一个大型的分布式摄像头网络，包括部署在支持Geotab的车辆上的移动的仪表盘摄像头，以及市政基础设施中的固定街道摄像头。这些数据的处理将依赖于最先进的机器学习方法，主要是深度卷积神经网络的变体。 该项目的成果将是开发基于深度学习的先进方法，以识别与道路安全相关的城市场景数据中的事件和特征，例如不安全的人行横道或高流量十字路口，或退化的市政基础设施，例如烧毁的路灯或坑洞。该项目的成果将包括通过设计一套结构化的实验，对这些方法的有效性进行全面的表征。此外，这些方法将在一个标准化的开发环境中实施，以便于转移到Geotab，随后进行商业化，并证明其适用于金斯顿市的市政基础设施监测。 该项目的影响将是推进机器学习方法的使用，以检测城市场景图像和视频中的安全关键事件和场景。该项目有可能改善城市环境的安全，既可以通过开发先进的驾驶员和/或行人警报，实时识别安全关键事件和条件，也可以通过告知道路和相关基础设施的设计标准，揭示某些事件的根本原因。该项目的进一步实际影响将是Geotab提供的新产品，以及为金斯顿市确定将这些技术应用于基础设施资产跟踪的机会。