ISSN 1991-2927
 

ACP № 2 (60) 2020

Author: "Ilia Vasilevich Voronov"

Sergei Vasilevich Voronov, Ulyanovsk State Technical University, Candidate of Engineering; graduated from the Faculty of Radioengineering of Ulyanovsk State Technical University; Associate Professor of the Radioengineering Department of UlSTU; an author of articles, monographs in the field of digital signal and image processing and computer vision. [e-mail: valmedia@yandex.ru]S. Voronov,

Rinat Nailevich Mukhometzianov, University of Waterloo, Canada, Postgraduate Student at the University of Waterloo, Canada; graduated from the Faculty of Radioengineering of Ulyanovsk State Technical University; an author of articles in the field of computer vision. [e-mail: mukhometzyanov@mail.ru]R. Mukhometzianov,

Ilia Vasilevich Voronov, Ulyanovsk State Technical University, Postgraduate Student at Ulyanovsk State Technical University; graduated from the Faculty of Radioengineering of Ulyanovsk State Technical University; an author of articles in the field of digital signal and image processing. [e-mail: ilvo1987@gmail.com]I. Voronov,

Vadim Andreevich Shramov, Federal Research-and-Production Center Joint Stock Company ‘Research-and-Production Association ‘Mars’, Postgraduate Student at Ulyanovsk State Technical University; graduated from the Faculty of Radioengineering of Ulyanovsk State Technical University; Research-Engineer at FRPC JSC ‘RPA ‘Mars’. [e-mail: vadim_shramov@mail.ru]V. Shramov

Trafic Sign Detection and Recognition in Real Time on Mobile Devices 52_13.pdf

An automatic traffic sign recognition system localizes road signs from images captured by an onboard camera of a vehicle and determine what road signs depicted. Such systems may support the driver on the road, be part of the self-driving cars or advanced driver assistance systems. This paper proposes an approach for solving traffic sign recognition problem using deep learning methods adopted to mobile devices with low power consumption. The approach consists of two consecutive stages: detection of a traffic sign and recognition of a class of the detected sign. Data for analysis were taken from three open sets of images. In order to analyze the effectiveness of the solution obtained, the results achieved were compared with the results of well-known approaches to object detection based on the use of deep convolutional neural networks. The results showed that the proposed algorithm provides the best recognition quality for all used data sets, as well as the highest recognition rate.

Deep learning, viola-jones detector, convolutional neural networks, traffic sign recognition, object detection.

2018_ 2

Sections: Artificial intelligence

Subjects: Artificial intelligence.


Sergei Vasilevich Voronov, Ulyanovsk State Technical University, Candidate of Engineering; graduated from the Faculty of Radioengineering of Ulyanovsk State Technical University; Associate Professor at the Department of Radioengineering; an author of articles and monographs in the field of digital signal and image processing and computer vision. [e-mail: valmedia@yandex.ru]S. Voronov,

Ilia Vasilevich Voronov, Ulyanovsk State Technical University, Postgraduate Student at Ulyanovsk State Technical University; graduated from the Faculty of Radioengineering of Ulyanovsk State Technical University; an author of articles in the field of digital signal and image processing. [e-mail: ilvo1987@gmail.com]I. Voronov,

Vadim Andreevich Shramov, Federal Research-and-Production Center Joint Stock Company ‘Research-and-Production Association ‘Mars’, Postgraduate Student at Ulyanovsk State Technical University; graduated from the Faculty of Radioengineering of Ulyanovsk State Technical University; Research Engineer at FRPC JSC “RPA “Mars”. [e-mail: vadim_shramov@mail.ru]V. Shramov

Detection of Objects on Images With the Use of Histograms of Orientated Gradients 000_9.pdf

Automated detection of objects of different classes on images and frames of video sequences is one of the main tasks of computer vision. The most widely used approach to solving this problem today is extracting certain features from the local image areas and further learning of classification algorithms on the basis of the extracted vectors. At the same time, the most interesting because of the good ratio of efficiency and the required computational resources are the features obtained using histograms of orientated gradients. This paper is devoted to the modification of the method of extraction of local features on the basis of histograms of orientated gradients, which consists in extracting features along the edges. It allows taking into account their spatial arrangement. Moreover, to increase the descriptive properties of feature vectors, it was suggested to use information on the structure that is based on the use of the "center of gravity" of local areas. Experimental results show that the proposed changes in comparison with the traditional method of extracting features allow to increase both the accuracy of detection of objects of different classes and the speed of convergence of classification algorithms.

Adaboost, object detection, hog, local features, adaboost, feature vector, classification.

2017_ 2

Sections: Information systems

Subjects: Information systems.


Aleksandr Grigorevich Tashlinskii, Ulyanovsk State Technical University, Doctor of Engineering, Professor, graduated from the Faculty of Radio-Engineering at Ulyanovsk Polytechnic Institute; Head of the Department of Radio-Engineering at Ulyanovsk State Technical University; an author of articles, monographs, and inventions in the field of digital processing of signals and images. [e-mail: tag@ulstu.ru]A. Tashlinskii,

Sergey Vasilievich Voronov, Ulyanovsk State Technical University, graduated from the Faculty of Radio-Engineering at Ulyanovsk State Technical University, Post-graduate student at the same University; an author of articles in the field of estimation of interframe geometric image deformations. [e-mail: valmedia@yandex.ru]S. Voronov,

Ilia Vasilevich Voronov, Ulyanovsk State Technical University, graduated from the Faculty of Radio-Engineering at Ulyanovsk State Technical University, Post-graduate student at the same University; an author of articles in the field of recurrent estimation of digital image parameters. [e-mail: ilvo1987@gmail.com]I. Voronov

Analysis of Objective Functions in a Problem of Estimation of Mutual Geometric Image Deformations 34_5.pdf

A convergence analysis of stochastic gradient estimation procedures without identification for the problem of the image mutual geometric deformation estimation is fulfilled. Interframe correlation coefficient, mean squared frame-to-frame difference and mutual information are considered as objective function

Image deformations’ estimation, objective function, correlation coefficient, mean squared frame-to-frame difference, mutual information.

2013_ 4

Sections: Mathematical modeling, calculi of approximation and software systems

Subjects: Mathematical modeling.


© FRPC JSC 'RPA 'Mars', 2009-2018 The web-site runs on Joomla!