In this work we introduce a pipeline to detect and recognize various utility meter numbers in the wild. The system leverages on deep neural networks for detection and recognition. In the detection phase, we employ a fully Convolutional Neural Network to perform a pixel-wise classification, while the recognition phase employs another deep neural network to predict the length and individual digits in a meter. We qualitatively showed that the proposed approach is robust against severe perspective distortions, different lighting conditions and blurred images. Furthermore, it is capable of detecting small scale digits. Our approach is suitable for billing companies aiming to increase efficiency, lowering the time consumed by manual checks performed in the billing process. Finally, we release the dataset used in this work to benchmark the task.

Reading Meter Numbers in the Wild

Alessandro Calefati;Ignazio Gallo
;
Shah Nawaz
2019-01-01

Abstract

In this work we introduce a pipeline to detect and recognize various utility meter numbers in the wild. The system leverages on deep neural networks for detection and recognition. In the detection phase, we employ a fully Convolutional Neural Network to perform a pixel-wise classification, while the recognition phase employs another deep neural network to predict the length and individual digits in a meter. We qualitatively showed that the proposed approach is robust against severe perspective distortions, different lighting conditions and blurred images. Furthermore, it is capable of detecting small scale digits. Our approach is suitable for billing companies aiming to increase efficiency, lowering the time consumed by manual checks performed in the billing process. Finally, we release the dataset used in this work to benchmark the task.
2019
2019 International Conference on Digital Image Computing: Techniques and Applications (DICTA)
9781728138572
International Conference on Digital Image Computing: Techniques and Applications (DICTA)
Perth, Australia
2-4 December 2019
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11383/2083817
 Attenzione

L'Ateneo sottopone a validazione solo i file PDF allegati

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 4
  • ???jsp.display-item.citation.isi??? 3
social impact