A Pipeline for Digital Meter Reading Based on YOLOv5 and PP-OCRv3

Intelligent recognition algorithm is the key to realize automatic reading of electrical meters in the power distribution room. In response to the difficulties such as complex recognition requirements and diverse character types of digital meters, we propose a meter reading pipeline, which includes the meter area detection by YOLOv5 and the character recognition by PP-OCRv3. At the same time, threshold screening of pixel values is proposed to identify the illuminated indicator lights of the digital meter. To verify the validity of the algorithm mentioned above, a digital meter detection and reading dataset containing more than 600 pictures is established based on real photography and data augmentation. Finally, the experimental results show that the method we propose can realize the real-time detection and accurate reading of digital meters, which is helpful to promote the automation, digitization and intelligence of distribution room inspection.