Technique for automatic classification of roads using the neural network Mask R-CNN

Abstract

Technique for automatic classification of roads using the neural network Mask R-CNN

Ignatyev A.V., Kulikov M.A., Tsapiev D.N., Tyrin V.V.

Incoming article date: 18.04.2022

The paper proposes a method for automatic classification of roads based on the use of a convolutional neural network Mask-R-CNN. The developed technique makes it possible to automate the task of categorizing roads, which is fundamental in the redistribution of traffic flows, since knowledge of the category of the road allows you to determine its maximum capacity. The article contains a description of the stages of training a neural network, as well as the results obtained when using it. The method of automatic road classification proposed in the paper showed good results both in classifying roads based on satellite images and in classifying roads based on photographs of road sections. When expanding the test set, the number of classes of recognized roads can be increased to match the categories of roads according to SP 34.13330.2021. In addition, this technique (in terms of segmenting objects in photographs) can be used to control the quality of the roadway.

Keywords: road categories, convolutional neural networks, satellite imagery, image segmentation, Mask R-CNN, image recognition, computer vision