Classification of road conditions from images with deep learning frameworks********

使用深度学习框架对图像中的路况进行分类********

• 批准号: 537911-2018
• 负责人: Ramanna, Sheela
• 金额: $ 1.82万
• 依托单位: University of Winnipeg
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2018
• 资助国家: 加拿大
• 起止时间: 2018-01-01 至 2019-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=662952
• 关键词: Classification; road; conditions; images; deep

This research project with Weatherlogics seeks to automate the process of observing highway conditions by using machine learning to automatically report road conditions using images taken by highway cameras. Specifically, Deep Convolutional Neural Networks (DCNNs) will be employed to classify images taken by highway cameras through the use of transfer learning. Modern DCNN architectures have attained remarkable success in image classification tasks. The deep learning algorithm will be able to take any static image of a highway and determine the road condition that is present (e.g. dry, wet, snow-covered, slush-covered, or ice-covered).****This project would have tremendous positive benefits for Canada. By more accurately observing road conditions, especially in winter, this technology would have both economic and social benefits. Due to the high skill involved in producing and running a machine learning model, the commercialization of this technology would provide high-quality models. Furthermore, the data from this model would be used to improve driver safety and reduce transportation impacts due to weather. Given the movement toward automation in vehicles, and intelligent transportation systems, the ability to receive real-time road condition information across the entire highway network would put Canada on the leading edge of these advances in transportation. Road weather forecasts and observations are a unique service that could be used to position Weatherlogics as a leading data provider to consumers, governments, and transportation/logistics companies that are impacted by adverse road weather****The significance of project is that this technology would give Weatherlogics a unique competitive advantage over traditional weather prediction companies.**

与 Weatherlogics 合作的这个研究项目旨在通过使用机器学习利用高速公路摄像机拍摄的图像自动报告道路状况，从而实现高速公路状况观察过程的自动化。具体来说，深度卷积神经网络（DCNN）将用于通过迁移学习对高速公路摄像机拍摄的图像进行分类。现代 DCNN 架构在图像分类任务中取得了显着的成功。深度学习算法将能够拍摄高速公路的任何静态图像并确定当前的道路状况（例如干燥、潮湿、积雪、雪泥覆盖或冰覆盖）。****该项目将为加拿大带来巨大的积极利益。 通过更准确地观察路况，特别是在冬季，该技术将具有经济效益和社会效益。由于生成和运行机器学习模型需要很高的技能，该技术的商业化将提供高质量的模型。此外，该模型的数据将用于提高驾驶员安全并减少天气造成的交通影响。鉴于车辆自动化和智能交通系统的发展，在整个高速公路网络中接收实时路况信息的能力将使加拿大处于交通运输进步的前沿。道路天气预报和观测是一项独特的服务，可用于将 Weatherlogics 定位为受恶劣道路天气影响的消费者、政府和运输/物流公司的领先数据提供商****该项目的意义在于该技术与传统天气预报公司相比，Weatherlogics 将具有独特的竞争优势。**