Pointer-Type Meter Recognition Based on Elliptical Bounding Box Detection and Reading Regression

The automatic reading of pointer-type meters plays an important role in routine inspection of power substation. However, considering various factors such as meter types, shooting angles, and lighting conditions, the recognition of the pointer meters in natural environment is still a challenging task. To address this issue, an improved YOLOv5 algorithm that using elliptical bounding box instead of traditional rectangle one is utilized to detect and classify dials, and a deep regression model based on UNet is designed to achieve meter readings. This algorithm has many advantages, such as without the need of pre-processing and post-processing, the effective filtering of background information through elliptical bounding boxes, thus can significantly improve the accuracy of reading recognition. Experimental results demonstrate that, within a reading error of ±5%, the recognition accuracy reaches 87.33%, while within a reading error of 10%, the recognition accuracy reaches 96.54%.