Deep Learning and Spatial Statistics for Determining Road Surface Condition

Machine Learning (ML), and especially Deep Learning (DL) methods, have evolved \nrapidly over the last years and showed remarkable advances in research areas such as computer \nvision and natural language processing; however, there are still engineering applications \nin industries such as transportation where DL methods have not been applied yet or that \ncan be benefited from an integrated approach using DL in addition to other methods. \nFor countries in Northern latitudes, one of such applications is Monitoring Road Surface \nCondition (RSC) during the Winter season for improving road safety and road maintenance \noperations. \nIn this study, we introduce a novel approach for monitoring of RSC that integrates \nDL methods and Spatial Statistics (SS) to simultaneously process data from roadside \ncameras and weather stations to determine automatically the category of snow coverage \nat sample locations across a region of interest. Our approach integrates the advantages of \nSS for interpolating spatial variables and the strengths of DL for Computer Vision tasks, \nparticularly for image classification. On one hand, SS models serve to understand the spatial \nautocorrelation of random variables and to determine their expected values in unsampled \nlocations based on a number of near observations. On the other hand, DL models extract \nrelevant patterns from a large number of training images and learn a mapping from input \nimages to a set of predefined labels. \nWe implement and evaluate our approach using data collected in the province of Ontario \nduring the 2017-2018 Winter season. Specifically, we included data from three separate \nsources, Environment Canada (EC) Weather stations, Road Weather Information System \n(RWIS) stations, and roadside cameras from the Ministry of Transportation of Ontario \n(MTO). To the best of our knowledge, this is the first study that integrates both DL and \nSS techniques for processing the three data sources with the goal of monitoring RSC. \nThe DL models we implement and compared are Inception, Inception-Resnet, Xception, \nDenseNet, MobileNetv2, and NASNet. All of these models have achieved remarkable \nresults for image classification in well-known benchmarks. The SS models we evaluate are \nOrdinary Kriging (OK), Radial Basis Functions (RBF), and Inverse Distance Weighted \n(IDW). The first provides a comprehensive understanding of the spatial autocorrelation \nfor each particular variable, while the second and third allow a faster implementation. Our \nintegrated approach works by combining the output feature vector from the DL model with \nthe interpolated values from the SS model to output a more robust prediction of RSC for \nthe locations of interest.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.