A Gauge Meter Reader Edge Device Based on Computer Vision and Deep Learning

Analog gauges are still used in industry and manufacturing sectors to gather analog values of one parameter to show the state of the machine or pipe. However, these devices are prone to human error readings and inaccurate samples. Therefore, this work aims to present an edge device with a small camera to digitalize analog samples from the analog gauge by using a novel pointer angle detector approach with an object-detection-vision technique. Yolo V5 nano was the target model and was quantized to be exported to the edge device to make inferences on it. As a main result, the system was tested in two analog gauges, one to present the atmospheric pressure (bar) and the other to show pressure measurement (psi). The system developed has an average ±0.261 bar deviation and a maximum error range of ±0.584 bar on one meter. In the case of another meter, it has an average ±1.438 psi deviation and a maximum error range of ±3.091 psi. The model developed weighs 2 MB.