It is aimed to detect natural gas meter indices by taking images with a mobile application.​

 This process normally occurs by meter reader personnel manually entering the numbers they see into a device. While making these readings, the staff reads hundreds of meters a day, and on the other hand, they have to enter the numbers correctly, take out the invoice and submit it under physically challenging conditions. As a result of these conditions, some incorrect readings may occur.  

 

The application basically detects the numbers by photographing the meter and processing the image from the image. It is designed to take images in quick succession for meter readers to use.

Detected index numbers and other information are displayed and saved in the database. Changes can be made to the determined number if desired. Records can be updated.

Barcode reader feature is available for meter number detection. The barcode on the meters can be read using the phone's camera if desired.

The inputs can be recorded together with their visuals and detections in a way that can be used for mass analysis later. These records can later be exported and the collected data can be evaluated.

 

 

The part of the captured photo containing the numbers is masked.

After pre-processing the masked image, detection is performed with a UNET model.​

 

 

The application has been tested in the field and worked with a 92% success rate.​

Thanks to the application, all meters read during the day can be recorded and followed up more clearly, error rates in readings can be reduced and time can be saved.

1. http://www.agaz.com.tr/koruklu-sayaclar/natek-koruklu-dogalgaz-sayaci/ ↩︎