A Novel Pointer Meter Recognition Scheme Based on Multitask Unified Network Architecture
IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 7505020-7505020
DOI: 10.1109/tim.2026.3704267 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Lifeng Zhang und 1 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben.
Abstract (Originalsprache)
Automatic pointer meter reading is a crucial technology for the Industrial Internet of Things (IIoT) and intelligent robot inspection. However, existing schemes struggle to balance accuracy and speed. Mainstream cascade models suffer from slow system inference speeds due to the sequential processing of detection and recognition networks. Furthermore, the widely used semantic segmentation techniques within these models introduce computational redundancy and impose high requirements on image quality. On the other hand, single-stage models are susceptible to noise interference in complex backgrounds, resulting in insufficient robustness. In addition, traditional rectification methods based on ellipse fitting have inherent defects, leading to systematic errors in readings under tilted views. To address these challenges, this article proposes a pointer meter recognition framework that achieves both high accuracy and high efficiency. First, we propose the Y-GaugeNet, a multitask unified model that performs meter localization, feature alignment, digit detection, and keypoint regression. This architecture retains the high-speed characteristic of single-stage models while possessing the local feature focusing capability of cascade models. Second, leveraging traditional image processing techniques, we design a training-free pointer position detection algorithm that achieves a single-module average inference speed exceeding 1000 FPS on CPU platforms with a mean relative error (MRE) of no more than 0.12%. Finally, a comprehensive theory of perspective transformation for tilted meters is established, which mathematically proves and resolves the issues present in traditional ellipse fitting methods. Final experimental results demonstrate that using an RTX 4060 GPU with an image resolution of $1024 \times 1024$ , our method achieves a balance between accuracy and speed, with an MRE of 0.17 % and an overall inference speed of 27 FPS, providing a reliable solution for real-time meter monitoring in industrial scenarios.
Zitieren
APA
Zhang, L., & Tang, J. (2026). A Novel Pointer Meter Recognition Scheme Based on Multitask Unified Network Architecture. IEEE Transactions on Instrumentation and Measurement, 75, 7505020-7505020. https://doi.org/10.1109/tim.2026.3704267
Vancouver
Zhang L, Tang J. A Novel Pointer Meter Recognition Scheme Based on Multitask Unified Network Architecture. IEEE Transactions on Instrumentation and Measurement. 2026;75:7505020-7505020. doi:10.1109/tim.2026.3704267
BibTeX
@article{Zhang2026,
author = {Zhang, Lifeng and Tang, Jianzhong},
title = {A Novel Pointer Meter Recognition Scheme Based on Multitask Unified Network Architecture},
journal = {IEEE Transactions on Instrumentation and Measurement},
year = {2026},
volume = {75},
pages = {7505020--7505020},
doi = {10.1109/tim.2026.3704267}
}
EUVIMED