Automatic Meter Reading Based on Bi-Fusion MSP Network and Carry-Out Rechecking

Abstract

Automatic meter reading is important for power billing in a smart city. Most SoTA (State-of-the-Art) vision-based methods can read only cyclometers and fail to handle dial meters due to their in-between problem and ambiguous patterns to interpret a digit and are not light enough to be run on an embedded platform. This paper focuses on the design and development of an Internet of Things (IoT)-assisted real-time Automatic Meter Reading (AMR) system for utility billing in a smart city. To enhance the accuracy of object detection, most SoTA methods use a very deep CNN-based architecture to create rich feature maps. However, this backbone also makes small objects in the last layer become one pixel or less. This paper proposes a novel BI-Fusion Mixed Stage Partial (BIF-MSP) network to hold the spatial information of a smaller object at the end of network architecture and also increase the efficiency while operating on an embedded system. It can accurately detect small digits not only from cyclometers but also from dial meters. It can automatically decide a rule (anticlockwise or clockwise) to accurately read digits on a dial-type meter. After that, a carry-out re-checking module is proposed to further improve the accuracy of this AMR system. The experiments show the superiorities of our ARM system in terms of accuracy and efficiency. The dataset can be publicly accessed from the following URL: https://140.113.110.150:5000/sharing/52HCvjly2

Keywords

• Automatic meter reading
• bi-fusion
• deep learning
• power billing
• object detection
• YOLO