A Robust Automatic Meter Reading System based on Mask-RCNN

Automatic Meter Reading (AMR) is tackled by leveraging the high capability of Region-Convolutional Neural Networks (RCNN). However, license plate recognition is quite popular, but AMR problem is still unexplored. Numerous investigations are still crucial due to some constraints: handcrafted features, low efficiency, blur, rotate digits, low reflection, and poor quality images. With the significance of robust AMR, we have achieved advances of AMR in computer vision and deep learning, but still, these methods lack of some errors. To address these problems, we proposed a method Mask-RCNN (AMR) based on mask region convolutional neural networks (Mask-RCNN) used for counter detection, digit segmentation, and recognition. Experiments demonstrated that our proposed method substantially outperforms the state-of-the-art methods in terms of counter detection and digit recognition on publically available UFPR-AMR dataset.