Modeling and implementation of the pavement detection module for automatic vehicle control using the U-NET neural network

Abstract

Modeling and implementation of the pavement detection module for automatic vehicle control using the U-NET neural network

Evsin V.A., Shirobokova S.N., Vorobyov S.P., Evsina V.A.

Incoming article date: 24.12.2021

This article discusses the problems of determining the road surface for automatic control of a vehicle using an artificial neural network. The current state of the industry is described, as well as the relevance of these studies. Describes the input data for determining the road surface. The idea of the applicability of the image segmentation method for determining the road surface is substantiated. The structure of an artificial neural network based on the U-NET architecture is being formed. In particular, the structure of the sequence of layers is described. Particular attention is paid to the mathematical modeling of the convolution process and the maximum pool. A mathematical model of the learning process of an artificial neural network, as well as activation functions: linear functions and sigmoids, is given. An algorithm for forming an artificial neural network model is proposed. The learning process of this function is visualized on the graph. The result of the training is presented.

Keywords: artificial neural networks, UNET, data analysis,, machine learning, deep lerning, convolutional neural networks, convolution, maximum pool, image segmentation, modeling