ML.
Projects
2022mobile

Smart Scanner

AI-powered electricity meter reading app for Sri Lanka

Mobile application using computer vision to automatically read Sri Lankan electricity meters. Built with Flutter and YOLOv2-tiny for on-device inference, with cloud-based YOLOv4 backend for enhanced accuracy.

Tech stack

  • Flutter
  • YOLOv2-tiny
  • TensorFlow Lite
  • AWS Lambda
  • YOLOv4
  • Python
Challenge
Accurately detecting and reading meter digits from various angles, lighting conditions, and meter types. The app needed to work offline for initial detection while offering enhanced cloud processing when online.
Solution
Trained custom YOLOv2-tiny model optimized for mobile deployment. Implemented dual-processing approach: on-device TFLite for instant results, cloud-based YOLOv4 on AWS Lambda for verification. Added image preprocessing pipeline to handle various lighting and angle conditions.
Impact
Reduced manual meter reading errors by 85%. Enabled utility companies to digitize meter readings with 92% accuracy, saving hours of manual data entry.