Deep Learning-Powered System for Real-Time Digital Meter Reading on Edge Devices

The ongoing reading process of digital meters is time-consuming and prone to errors, as operators capture images manually update the system with new readings. This work proposes automate this operation through a deep learning-powered solution for universal controllers flow that can be seamlessly incorporated into operators’ existing workflow. Firstly, display area equipment extracted screen detection module, perspective correction step performed. Subsequently, text regions are identified fine-tuned EAST detector, important readings selected template matching based on expected graphical structure. Finally, convolutional recurrent neural network model recognizes registers it. Evaluation experiments confirm robustness potential workload reduction proposed system, which correctly extracts 55.47% 63.70% values in controllers, 73.08% from meters. Furthermore, pipeline performs real time low-end mobile device, an average execution preview under 250 ms acquired photo 1500 entire pipeline.