A pointer meter reading recognition method based on YOLOX and semantic segmentation technology

Pointer meters are extensively employed in power, chemical, automotive, and other modern production processes. To improve the accuracy and robustness of the existing reading recognition algorithms for pointer meters, this research proposed a novel approach based on YOLOX convolutional network and semantic segmentation technology. Firstly, the dial of the target meter is detected from the background image using the YOLOX network. Secondly, the meter's main tick marks, dial center, and pointer are determined by the proposed semantic segmentation solution using the attention U-Net network and then corrected the tilt through perspective transformation. Finally, the accurate reading is achieved by using the enhanced angle method that calculates the reading value using the angles between the pointer and the two most adjacent main tick marks. A series of experiments have been conducted to evaluate the presented approach's feasibility and robustness. The results indicated that the fiducial errors of its reading values are no more than 0.31%.

Automated summary

A novel approach based on YOLOX convolutional network and semantic segmentation technology is proposed to improve the accuracy and robustness of reading recognition algorithms for pointer meters. The approach detects the dial of the target meter using the YOLOX network, determines the meter's main tick marks, dial center, and pointer through semantic segmentation, corrects the tilt through perspective transformation, and calculates the reading value using the angles between the pointer and main tick marks. Experimental results show that the presented approach achieves reading values with fiducial errors of no more than 0.31%.