GAS meter reading from real world images using a multi-net system

Vanetti, Marco; Gallo, Ignazio; Nodari, Angelo

doi:10.1016/j.patrec.2012.11.014

Cited by 35 publications

(46 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Counter detection is evaluated in the UFPR-AMR dataset and also in two public datasets, 5,13 while counter recognition is assessed only in the UFPR-AMR dataset. This is because (i) two different sets of images were used to evaluate digit segmentation and recognition in Ref.…”

Section: Resultsmentioning

confidence: 99%

“…In Refs. 5 AMR presents an unusual challenge in OCR: rotating digits. Typically, this is the major cause of errors, even when robust approaches are employed for digit recognition.…”

Section: Related Workmentioning

confidence: 99%

“…It is important to note that in the dataset introduced in Ref. 5,153 images are divided into different subsets for the evaluation of each stage and only 100 images are for the end-to-end evaluation of the AMR system. Also, there is no split protocol in Ref.…”

Section: Datasetsmentioning

confidence: 99%

“…2,5 Since this operation is prone to errors, another operator needs to check the proof image to confirm the reading. 2,5 This offline checking is expensive in terms of human effort and time, and has low efficiency. 6 Moreover, due to a large number of images to be evaluated, the inspection is usually done by sampling 7 and errors might go unnoticed.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Convolutional neural networks for automatic meter reading

et al. 2019

View full text Add to dashboard Cite

In this paper, we tackle Automatic Meter Reading (AMR) by leveraging the high capability of Convolutional Neural Networks (CNNs). We design a two-stage approach that employs the Fast-YOLO object detector for counter detection and evaluates three different CNN-based approaches for counter recognition. In the AMR literature, most datasets are not available to the research community since the images belong to a service company. In this sense, we introduce a new public dataset, called UFPR-AMR dataset, with 2,000 fully and manually annotated images. This dataset is, to the best of our knowledge, three times larger than the largest public dataset found in the literature and contains a well-defined evaluation protocol to assist the development and evaluation of AMR methods. Furthermore, we propose the use of a data augmentation technique to generate a balanced training set with many more examples to train the CNN models for counter recognition. In the proposed dataset, impressive results were obtained and a detailed speed/accuracy trade-off evaluation of each model was performed. In a public dataset, state-of-the-art results were achieved using less than 200 images for training.

show abstract

Section: Resultsmentioning

confidence: 99%

“…In Refs. 5 AMR presents an unusual challenge in OCR: rotating digits. Typically, this is the major cause of errors, even when robust approaches are employed for digit recognition.…”

Section: Related Workmentioning

confidence: 99%

Section: Datasetsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Convolutional neural networks for automatic meter reading

et al. 2019

View full text Add to dashboard Cite

show abstract

“…Com respeito ao processamento das imagens de medidores, os trabalhos apresentam o emprego de variadas técnicas, dentre as quais a transformação da imagem em tons de cinza (EDWARD, 2013), segmentação por limiarização (ALEXANDRIA, 2004;ZHAO et al, 2009), morfologia matemática (LEI et al, 2010 e rotulação de componentes conexos (OLIVEIRA et al, 2009). Para o reconhecimento dos dígitos, técnicas como Redes Neurais Artificiais (RNA) VANETTI et al, 2012), classificação baseada na distância mínima (NAVA-ORTIZ et al, 2011), correlação cruzada (BELAN et al, 2013), entre outras, são utilizadas. São tomadas como base diversas características das imagens dos dígitos, tais como histograma de projeções de pixels (CASTELLS-RUFAS; CARRABINA, 2006), tipo de traços retos ou curvos no dígito (SHU et al, 2007), momentos invariantes (BAI et al, 2010), dentre outras.…”

Section: Trabalhos Relacionadosunclassified

Utilização de Técnicas de Processamento de Imagens e Classificação de Padrões no Reconhecimento de Dígitos em Imagens de Medidores de Consumo de Gás Natural

Gonçalves

Centeno

2017

Abakos

View full text Add to dashboard Cite

ResumoEste trabalho propõe uma abordagem que emprega técnicas de processamento de imagens e classificação de padrões para o reconhecimento de dígitos apresentados no contador de consumo de medidores de gás natural. Tais imagens são obtidas em campo a partir de condições reais de operação, diferentemente da maioria das abordagens encontradas na literatura, que se baseiam em imagens adquiridas em ambientes controlados. Inicialmente, o contador de consumo é segmentado por técnicas de processamento de imagens. A segmentação é realizada com base no espaço de cor HSL da imagem, diferentemente da maioria dos trabalhos apresentados na literatura, que utilizam imagens em tons de cinza. Em seguida, os dígitos são individualmente segmentados e suas características são extraídas de modo a compor uma base de conhecimento. Essa base serve de apoio para realizar a classificação e o reconhecimento dos dígitos. A metodologia mostrou-se promissora nesse cenário, chegando a alcançar 95% de taxa de acerto no reconhecimento dos dígitos. Apresentou menos de 5% de falha no processo de segmentação do contador de consumo, considerando-se uma base com 903 imagens de medidores de gás.

show abstract