Принципи організації нейромережної системи розпізнавання символів номерних знаків
Вантажиться...
Файли
Дата
2018
Автори
Лаврінчук, В.В.
Чешун, В.М.
Чорненький, В.І.
Lavrinchuk, V.V.
Cheshun, V.M.
Chornenkyi, V.I.
Назва журналу
Номер ISSN
Назва тому
Видавець
Хмельницький національний університет
Анотація
В роботі запропоновано алгоритмічні, функціональні та структурні рішення для побудови економічної
та високопродуктивної системи автоматизованого розпізнавання номерних знаків автомобільного транспорту,
що досягається за рахунок використання в її складі реалізованої апаратно на програмованих логічних
інтегральних схемах тришарової штучної нейронної мережі прямого поширення, орієнтованої на виконання
функцій розпізнавання символів номерного знаку за сегментами зображень фіксованої розмірності.
Due to the constant increase in the number of cars on the roads in Ukraine, the problem of automation in control and maintenance of traffic participants is exacerbated. In the article the peculiarities of existing systems and methods of license plates recognition are analyzed, their main advantages and disadvantages are determined. Execution of a character recognition operation by the program reduces system performance or requires the use of expensive computing systems. To create a high-performance recognition system, it is proposed to use the hardware-implemented artificial neural network. The basis of the system is a three-layer artificial neural network of direct passage, which is implemented on programmable logic integrated circuits and performs the functions of recognizing license plate characters from fixed-size image segments. Allocation of a license plate in a photograph and segmentation are carried out by traditional methods. Converting the selected segments to the desired size is implemented before the character recognition operation. The neural network receives a binary vector, in which each bit corresponds to one pixel of the image segment. The article substantiates the choice of the size of the segments and proposes an algorithm, functional and structural solutions for building an improved system of automatic identification of license plates. The method of error propagation is used to calculate the parameters of the artificial neural network. The simulation of the system yielded 96.81% of the true recognition results. The given solutions are the basis for the hardware implementation of the artificial neural network on programmable logic integrated circuits, which should provide the possibility of creating a high-performance recognition system at a moderate cost.
Due to the constant increase in the number of cars on the roads in Ukraine, the problem of automation in control and maintenance of traffic participants is exacerbated. In the article the peculiarities of existing systems and methods of license plates recognition are analyzed, their main advantages and disadvantages are determined. Execution of a character recognition operation by the program reduces system performance or requires the use of expensive computing systems. To create a high-performance recognition system, it is proposed to use the hardware-implemented artificial neural network. The basis of the system is a three-layer artificial neural network of direct passage, which is implemented on programmable logic integrated circuits and performs the functions of recognizing license plate characters from fixed-size image segments. Allocation of a license plate in a photograph and segmentation are carried out by traditional methods. Converting the selected segments to the desired size is implemented before the character recognition operation. The neural network receives a binary vector, in which each bit corresponds to one pixel of the image segment. The article substantiates the choice of the size of the segments and proposes an algorithm, functional and structural solutions for building an improved system of automatic identification of license plates. The method of error propagation is used to calculate the parameters of the artificial neural network. The simulation of the system yielded 96.81% of the true recognition results. The given solutions are the basis for the hardware implementation of the artificial neural network on programmable logic integrated circuits, which should provide the possibility of creating a high-performance recognition system at a moderate cost.
Опис
Ключові слова
система розпізнавання символів, номерні знаки, штучні нейронні мережі, програмовані логічні інтегральні схеми, character recognition system, license plates, artificial neural networks, programmable logic integrated circuits
Бібліографічний опис
Лаврінчук, В.В. Принципи організації нейромережної системи розпізнавання символів номерних знаків [Текст] / В. В. Лаврінчук, В. М. Чешун, В. І. Чорненький // Вісник Хмельницького національного університету. Технічні науки. – 2018. – № 1. – С. 48-53.