Intelligent System for Recognition of Road Infrastructure Objects Using a Neural Network Model

Abstract

The article deals with the problem of recognition of road infrastructure objects. The expediency of solving this problem using an intelligent system based on a neural network model is updated. In order to develop it to solve the problem, the collection of initial data, the choice of libraries, the construction of the architecture and the training of the neural network were carried out. The initial data for constructing the model was obtained when generating queries in a search engine and presented in the form of images of traffic signs, road markings and traffic lights. To build a neural network, training and test samples were formed for 3 classes of images, taking into account various road conditions. When building a neural network model, the activation functions Relu and Softmax were used. The neural network model includes 4 convolution layers, 2 subsampling layers and 2 fully connected layers. The IDLE development environment was used to train the model. The program code is written in the Python 3.9.7 programming language. The neural network model was trained for 10 epochs. An example of the operation of the system and the results of testing the developed built model are presented. Studies have shown that the developed intelligent system and neural network model for recognizing road infrastructure objects are adequate, and its accuracy is 92 %. The achieved accuracy of the neural network model is not inferior to the accuracy of other methods for classifying objects in images.