Development of image recognition system for analogue meter’s reading detection

Chong, Yue Jiet (2022) Development of image recognition system for analogue meter’s reading detection. Final Year Project, UTAR.

Preview

PDF Download (2779Kb) | Preview

Abstract

Digitization is one of the major components in Industrial Revolution 4 (IR 4.0). It provides a lot of benefits to the industrial, such as increasing productivity, better data visualisation, simplifying parameters control, etc. Although many meters nowadays come with the smart Internet of Things (IoT) features, which provides real-time monitoring and data storing, many industries still prefer to continue using the existing analogue meters in their manufacturing plants as replacing the existing analogue meters with the cloud-connected digital meters can be very costly especially for industrial grade meters. In this project, a costeffective image recognition system to capture and digitize the analogue meter’s readings using deep learning model (SSD MobileNet) as well as optical character recognition (Tesseract) was demonstrated. The deep learning model has been trained with a dataset of 750 images and was used to detect the region of interest (meter’s readings). The OCR is used to convert the readings to string datatype. Besides, the image processing techniques via OpenCV library has been implemented for enhancing the quality of the ROI. The programme developed has been transferred and executed on the Raspberry Pi microcomputer with camera module attached to an analogue water meter. The results show that the accuracies of the deep learning model and OCR are 95% and 91%, respectively. In addition, the memory occupation of the deep learning model is about 10 MB, which is suited the embedded system with limited memory capacity.

Actions (login required)