A Pointer Meter Reading Recognition Method Based on Line Detection and Keypoint Localization
Gaoxin Wu, Qiugan Huang, Wangming Xu · 2025
Analog pointer meters have been widely used in industrial settings for monitoring critical equipment and ensuring operational safety. However, the automatic reading recognition for analog pointer meters under complex environmental conditions remains a challenging task, as accurately locating the pointer position is often affected by various interfering factors in practical applications. In this work, we propose a pointer meter reading recognition method that combines pointer line detection and dial keypoint localization. First, a deep Hough transform-based Pointer Line Detection Network (PLNet) is developed by integrating the Hough transform into a convolutional neural network to detect the pointer lines. Then, a Pointer Direction Prediction Network (PDNet) is introduced to determine the direction of the pointer. Next, the YOLOv8s-pose network is employed to locate the scale numbers and the tick points on the dial. Finally, the reading can be calculated based on the positional relationships between the pointer line and the tick points through a series of post-processing steps. Experimental results on a self-constructed multi-type pointer meter dataset demonstrate the effectiveness of the proposed method.