1,720,964 research outputs found
An efficient and layout-independent automatic license plate recognition system based on the Yolo detector
Full text linkOrientador: Prof. Dr. David Menotti GomesDissertação (mestrado) - Universidade Federal do Paraná, Setor de Ciências Exatas, Programa de Pós-Graduação em Informática. Defesa : Curitiba, 08/03/2019Inclui referências: p. 83-92Área de concentração: Ciência da ComputaçãoResumo: O reconhecimento automático de placas de veículos (Automatic License Plate Recognition (ALPR), do inglês Automatic License Plate Recognition) tem sido um tópico frequente de pesquisa devido as muitas aplicações praticas tais como cobrança automática de pedágio e aplicação da lei de transito. No entanto, muitas das soluções atuais ainda não são robustas em situações do mundo real, dependendo comumente de certas restrições como câmeras ou ângulos de visão específicos, planos de fundo simples, boas condições de iluminação, entre outras. Esta dissertação apresenta um sistema ALPR eficiente e independente de layout baseado no detector de objetos de ultima geração YOLO (You Only Look Once), com uma abordagem unificada para detecção de placas e classificação de layout para melhorar os resultados de reconhecimento através de regras de pós-processamento. Em cada estagio, nos avaliamos diferentes modelos com varias modificações, otimizando e combinando-os cuidadosamente com o objetivo de alcançar o melhor compromisso de velocidade/precisão. As Redes Neurais Convolucionais (Convolutional Neural Networks (CNNs), do inglês Convolutional Neural Networks) são treinadas utilizando imagens de vários conjuntos de dados para que sejam robustas sob diferentes condições (por exemplo, com variações de iluminação, posição e configurações da câmera, tipos de veículos, etc.). Este trabalho também introduz um conjunto de dados publico para ALPR, chamado UFPR-ALPR, que inclui 4.500 imagens totalmente anotadas de 150 veículos em cenários do mundo real em que tanto o veiculo quanto a câmera (dentro de outro veiculo) estão em movimento. Em comparação com o conjunto de dados publico de placas brasileiras mais empregado para ALPR, o conjunto de dados proposto tem mais que o dobro de imagens e contem uma variedade maior em diferentes aspectos. O sistema proposto foi capaz de atingir uma taxa media de reconhecimento de ponta a ponta de 96,76% em oito conjuntos de dados públicos utilizados nos experimentos, superando tanto os trabalhos anteriores quanto os sistemas comerciais nos conjuntos de dados ChineseLP, OpenALPR-EU, SSIG e UFPR-ALPR. Nos demais conjuntos de dados, a abordagem proposta obteve resultados semelhantes ao melhor resultado alcançado pelas linhas de base. Nosso sistema também alcançou impressionantes taxas de quadros por segundo (FPS, do inglês Frames Per Second) em uma unidade de processamento gráfico (GPU, do inglês Graphics Processing Unit) de ponta, sendo capaz de executar em tempo real mesmo quando ha 4 veículos na cena.Abstract: ALPR has been a frequent topic of research due to many practical applications such as automatic toll collection and traffic law enforcement. However, many of the current solutions are still not robust in real-world situations, commonly depending on certain constraints such as specific cameras or viewing angles, simple backgrounds, good lighting conditions, among others. This dissertation presents an efficient and layout-independent ALPR system based on the state-of-the-art You Only Look Once (YOLO) object detector, with a unified approach for License Plate (LP) detection and layout classification to improve the recognition results through post-processing rules. In each stage, we evaluate different models with various modifications, carefully optimizing and combining them aiming to achieve the best speed/accuracy trade-off. The CNNs are trained using images from several datasets so that they are robust under different conditions (e.g., with variations in lighting, camera position and settings, vehicle types, etc.). This work also introduces a public dataset for ALPR, called UFPR-ALPR, that includes 4,500 fully annotated images from 150 vehicles in real-world scenarios where both the vehicle and the camera (inside another vehicle) are moving. Compared to the public dataset of Brazilian LPs most frequently used for ALPR, our dataset has more than twice the images and contains a larger variety in different aspects. The proposed system was able to achieve an average end-to-end recognition rate of 96.76% across eight public datasets used in the experiments, outperforming both previous works and commercial systems in the ChineseLP, OpenALPR-EU, SSIG and UFPR-ALPR datasets. In the other datasets, the proposed approach obtained similar results to the best result attained by the baselines. Our system also achieved impressive Frames Per Second (FPS) rates on a high-end Graphics Processing Unit (GPU), being able to perform in real time even when there are 4 vehicles in the scen
Automatic license plate recognition (ALPR) : toward improving the state of the art and bridging the gap between academia and industry
Full text linkOrientador: David MenottiCoorientador: Rodrigo MinettoTese (doutorado) - Universidade Federal do Paraná, Setor de Ciências Exatas, Programa de Pós-Graduação em Informática. Defesa : Curitiba, 25/04/2024Inclui referênciasÁrea de concentração: Ciência da ComputaçãoResumo: O reconhecimento automático de placas de veículos (ALPR) tem sido um tópico de pesquisa frequente devido às suas amplas aplicações práticas, incluindo cobrança automática de pedágios e aplicação das leis de trânsito. Apesar do progresso considerável no estado da arte nos últimos anos, várias questões persistem em aberto neste domínio. Esta tese investiga o potencial para avanços significativos no ALPR ao investigar e abordar meticulosamente essas questões, em vez de focar no aumento do número de imagens reais de treinamento, na proposta de descritores inovadores, ou na busca extensiva por melhores arquiteturas de modelos. Nossa pesquisa começa endereçando a falta de atenção dada às imagens contendo placas Mercosul, motocicletas, e placas com duas linhas de caracteres através da criação de um conjunto de dados dedicado (RodoSol-ALPR) e da condução de uma série de experimentos com ele. Nossos experimentos ressaltam a importância deste conjunto de dados para o reconhecimento robusto de placas Mercosul e de placas com duas linhas de caracteres, já que modelos de reconhecimento óptico de caracteres (OCR) treinados em outros conjuntos de dados não conseguem ultrapassar uma taxa de reconhecimento de 70% em seu conjunto de teste. Posteriormente, apresentamos melhorias substanciais no desempenho do ALPR de ponta a ponta ao mesclar a saída de vários modelos de OCR e combinar várias metodologias de geração de dados sintéticos. Notavelmente, a utilização extensiva de dados sintéticos leva a resultados estado-da-arte em diversos conjuntos de dados e desempenha um papel fundamental na superação de desafios causados pela disponibilidade limitada de dados de treinamento. Esta tese também identifica questões críticas na avaliação de sistemas para o ALPR. Revelamos que os protocolos de avaliação estabelecidos não levam em conta as quase duplicatas nos conjuntos de treinamento e teste, dificultando o desenvolvimento e a aceitação de modelos mais eficientes que tenham fortes habilidades de generalização mas não memorizam duplicatas tão bem quanto outros modelos. Por fim, contextualizamos o problema do viés de conjunto de dados no domínio do ALPR, aumentando a conscientização sobre suas possíveis consequências. A identificação destas questões enfatiza a importância da realização de experimentos cross-dataset, uma vez que estes fornecem uma melhor indicação de generalização do que experimentos intra-dataset. Uma maior adoção de avaliações cross-dataset tem o potencial de reduzir a lacuna entre os resultados relatados no meio acadêmico e os alcançados na indústria.Abstract: Automatic License Plate Recognition (ALPR) has been a frequent research topic due to its wideranging practical applications, including automatic toll collection and traffic law enforcement. Despite the considerable progress in the state of the art driven by deep learning and the increasing availability of public datasets, several open issues persist within the ALPR domain. This thesis investigates the potential for significant advancements in ALPR by meticulously identifying and addressing these issues, rather than focusing on increasing the number of real training images, designing groundbreaking descriptors, or extensively searching for better model architectures. Our research begins by tackling the lack of attention given to images featuring Mercosur License Plates (LPs), motorcycles, and two-row LPs by creating a dedicated dataset (RodoSol-ALPR) and conducting a series of experiments using it. Our experiments underscore the importance of the RodoSol-ALPR dataset for robust recognition of Mercosur and two-row LPs, as Optical Character Recognition (OCR) models trained on alternative datasets fail to surpass a 70% recognition rate on its test set. Subsequently, we showcase substantial improvements in end-to-end ALPR performance by fusing the outputs of multiple OCR models and combining various synthetic data generation methodologies. Notably, the extensive use of synthetic data leads to state-of-the-art results across diverse datasets and plays a pivotal role in overcoming challenges caused by limited training data availability. This thesis also identifies critical issues in the assessment of ALPR systems. We reveal that established evaluation protocols have failed to account for near-duplicates within training and test sets, hindering the development and acceptance of more efficient models that have strong generalization abilities but do not memorize duplicates as well as other models. Finally, we contextualize the dataset bias problem within the License Plate Recognition (LPR) domain, raising awareness about its potential consequences and discussing the subtle ways this bias may have crept into existing datasets. Identifying these issues emphasizes the importance of conducting cross-dataset experiments, as they provide a better indication of generalization than intra-dataset ones. This shift toward cross-dataset setups has the potential to bridge the gap between results reported in academia and those achieved in industry
Going Beyond Counting First Authors in Author Co-citation Analysis
Full text linkThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Full text link“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Appropriate Similarity Measures for Author Cocitation Analysis
Full text linkWe provide a number of new insights into the methodological discussion about author cocitation analysis. We first argue that the use of the Pearson correlation for measuring the similarity between authors’ cocitation profiles is not very satisfactory. We then discuss what kind of similarity measures may be used as an alternative to the Pearson correlation. We consider three similarity measures in particular. One is the well-known cosine. The other two similarity measures have not been used before in the bibliometric literature. Finally, we show by means of an example that our findings have a high practical relevance.information science;Pearson correlation;cosine;similarity measure;author cocitation analysis
Dispelling the Myths Behind First-author Citation Counts
Full text linkWe conducted a full-scale evaluative citation analysis study of scholars in the XML research field to explore just how different from each other author rankings resulting from different citation counting methods actually are, and to demonstrate the capability of emerging data and tools on the Web in supporting more realistic citation counting methods. Our results contest some common arguments for the continued
use of first-author citation counts in the evaluation of scholars, such as high correlations between author rankings by first-author citation counts and other citation
counting methods, and high costs of using more realistic citation counting methods that are not well-supported by the ISI databases. It is argued that increasingly available digital full text research papers make it possible for citation analysis studies to go beyond what the ISI databases have directly supported and to employ more
sophisticated methods
koamabayili/VECTRON-author-checklist: VECTRON author checklist
No full textWe have done our best to complete the author checklist relating to the use of animals in the hut study. Note that the objective for the hut study was to evaluate the IRS treatment applications for residual efficacy against Anopheles mosquitoes, including the local An. coluzzii mosquito population. Cows were only used to attract mosquitoes into the huts and no tests were carried out directly on the cows. The author checklist is intended for use with studies where experiments are carried out on animals, which is why we have had such difficulty in completing this for the hut study, as many of the questions do not relate to how the cows were used
Author-wise bibliometric analysis based on entropy.
No full textAuthor-wise bibliometric analysis based on entropy.</p
- …
