MASc Seminar: Deep Learning and Spatial Statistics for determining Winter Road Surface Condition

Candidate: Juan Manuel Carrillo Garcia

Title: Deep Learning and Spatial Statistics for determining Winter Road Surface Condition

Date: May 30, 2019

Time: 3:00pm

Place: E5  4128

Supervisor(s): Crowley, Mark

Abstract:

Road maintenance during the Winter season is a safety critical and resource demanding operation. One of its key activities is determining road surface condition (RSC) in order to prioritize roads and allocate cleaning efforts such as plowing or salting. Two conventional approaches for determining RSC are:

visual examination of roadside camera images by trained personnel and patrolling the roads to perform on-site inspections. However, with more than 500 cameras collecting images across Ontario, visual examination becomes a resource-intensive activity, difficult to scale especially during periods of snow storms. This thesis makes two novel contributions to improve the efficiency of Winter road maintenance operations. To the best of our knowledge, this is the first study that combines Deep Learning methods along with Spatial Statistics for an Engineering application in the transportation sector.

First, we use multiple Deep Learning models to automatically determine RSC from roadside camera images and weather variables, extending previous research where similar methods have been used to deal with the problem. The dataset we used was collected during the 2017-2018 Winter season from 40 stations connected to the Ontario Road Weather Information System (RWIS), it includes 14.000 labeled images and 70.000 weather measurements. In particular, we train and evaluate the performance of seven state-of-the-art models from the Computer Vision literature, including the recent DenseNet, NASNet, and MobileNet. Also, by integrating observations from weather variables, the models are able to better ascertain RSC under poor visibility conditions.

Additionally, we suggest the integration of data collected from the RWIS stations with images from other MTO (Ministry of Transportation of Ontario) roadside cameras and weather data from Environment Canada stations. Using spatial statistics we quantify the benefits of integrating the three datasets across Southern Ontario, showing evidence of a six-fold increase in the number of available roadside cameras and therefore improving the spatial coverage in the most populous ecoregions in Ontario. Moreover, we evaluate three spatial interpolation methods for inferring weather variables in locations without weather measurement instruments and identify the one that offers the best tradeoff between accuracy and ease of implementation.