Devit Technology developed a cutting-edge Automatic Meter Reading (AMR) system using advanced OCR technology for the City of Tshwane. This innovative solution combines computer vision, deep learning, and mobile technology to automatically scan and interpret South African municipal utility meters with state-of-the-art accuracy.
AI OCR System
City of Tshwane AMR System
Overview
Technologies Used
Python
YOLOv5
Pytorch
TensorFlow
Android
Java
Ionic
React
TypeScript
ElectronChallenge
Manual meter reading was time-consuming, error-prone, and required significant human resources. The City of Tshwane needed an automated solution that could accurately read various meter types, work in different lighting conditions, and integrate with existing municipal systems while reducing operational costs.
Approach
We developed a comprehensive AI-powered solution combining multiple technologies. Our approach included creating a custom data annotation platform to speed up model training, implementing YOLOv5 for meter detection, and developing CRNN models for text recognition. The system was deployed as a native Android application with Cordova and Capacitor plugins for seamless hardware integration.
Key Features
01
Advanced computer vision for meter detection
02
Deep learning text recognition (CRNN)
03
Native Android application
04
Cordova and Capacitor plugin integration
05
SAP server integration
06
Real-time data processing
07
Offline capability with sync
08
Multi-meter type support
09
Auto-inference data annotation
10
Training data collection and management
Results
The AI OCR system achieved over 95% accuracy in meter readings across various meter types and lighting conditions. Processing time was reduced from hours to minutes, and the system successfully integrated with SAP data capture servers for seamless data flow.
Impact
Reduced meter reading costs and improved accuracy. The system can process thousands of meter readings daily with minimal human intervention.