Even though smart instruments are popular in modern industrial processes, pointer meters without computer interfaces are still widely employed in many application areas. Automatic reading recognition of pointer meters based on computer vision provides a promising solution to input the meter readings to the existing monitoring system or IoT. However, most of the current reading recognition approaches for pointer meters are based on inconvenient wired systems or expensive inspection robots. Compared with a wired system, wireless sensor networks (WSNs) are low-cost, easy to install and relocate. To leverage these advantages of WSNs and address the limitation of its radio bandwidth, this paper proposes a novel automated meter reading system for pointer meter using WSNs with on-sensor image processing, in which meter image capturing, image preprocessing, and reading recognition are completed on WSNs end node, and then only the recognition result is transmitted through WSNs. The feasibility of the proposed method is evaluated by a set of experiments on the prototype. Experimental results indicate that the presented approach can recognize the meter reading with an error of 0.3% and significantly reduce the payload transmission data in the WSNs.

尽管智能仪表在现代工业过程中很流行，但没有计算机接口的指针式仪表仍广泛应用于许多应用领域。基于计算机视觉的指针式仪表自动读数识别为将仪表读数输入现有监控系统或物联网提供了一种很有前景的解决方案。然而，目前大多数指针式仪表的读数识别方法都是基于不方便的有线系统或昂贵的检查机器人。与有线系统相比，无线传感器网络 (WSN) 成本低、易于安装和迁移。为了利用 WSN 的这些优势并解决其无线电带宽的限制，本文提出了一种新颖的指针式仪表自动抄表系统，使用带有传感器图像处理的 WSN，其中仪表图像捕获，图像预处理和阅读识别在WSNs端节点完成，然后只有识别结果通过WSNs传输。通过对原型的一组实验来评估所提出方法的可行性。实验结果表明，所提出的方法可以以 0.3% 的误差识别抄表，并显着减少 WSN 中的有效载荷传输数据。