<![CDATA[Manual recording of water meters in the Community-Based Drinking Water and Sanitation Program (PAMSIMAS) is still vulnerable to human error, reporting delays, and operational inefficiencies. This study proposes a smartphone-based Automated Meter Reader (AMR) system that utilizes Optical Character Recognition (OCR) as a low-cost digital solution for rural environments. The system uses a smartphone camera to capture an analog water meter image and processes it through a computer vision pipeline that includes grayscale conversion, bilateral filtering, Canny edge detection, contour-based segmentation, and a full-image OCR fallback mechanism. An experimental evaluation was conducted on 120 analog water meter images with variations in lighting, capture angle, blur level, and meter surface conditions. Digit extraction was performed using Google Vision OCR (online) and Tesseract OCR (offline). The OCR accuracy was calculated based on the compatibility of the digit value of the recognition result with the ground truth value and complemented by confidence score analysis. The test results showed an average OCR accuracy of 91%, a confidence score of 0.87, and an average processing time of 1.27 s per image. Although the system showed stable performance in most test scenarios, the accuracy declined in strong glare conditions and with faulty meters, indicating the limitations of the contour-based segmentation approach. Overall, this smartphone-based AMR system has proven to be feasible and practical for supporting the digitization of community-based water management, with the potential for further development through deep learning-based segmentation.]]>
Copyrights © 2025
