<![CDATA[Background: The main challenge in this research is the vulnerability of manual gauge readings to human error, particularly under degraded visual conditions such as lighting distortion, tilted perspective, and blurred images. Furthermore, achieving fully automatic gauge reading remains challenging because many existing approaches still depend on manually predefined scale ranges, perspective-correction preprocessing, or limited gauge configurations. Objective: This study aims to automate the reading of analog gauges, which are widely used across various industrial sectors. Methods: This study proposes an automatic analog gauge-reading framework based on YOLOv9, SAM, and TrOCR. YOLOv9 was used to detect the face of the analog gauges and their components in the image, followed by prompt-guided SAM segmentation to extract the major and minor needle and scale ticks. Meanwhile, the scale-number recognition is performed by the TrOCR. Unlike several previous approaches, the proposed method calculates gauge values through ellipse-based angular mapping without adding specific steps for prospective correction and manually providing initial scale ranges. Then, this approach was evaluated using datasets consisting of analog gauge images from various sources and representing several instrument types. Result: The proposed system achieved an average relative error (ARelE) of 1.935% and an average reference error (ARefE) of 0.601%. Furthermore, the robustness evaluation of this framework yielded stable performance across several degraded visual variations. Conclusion: The main contribution of this study is the integration of PGS, transformed-based OCR, and ellipse-based interpretation into a single pipeline for adaptive analog gauge reading. This study presented potential industrial monitoring solutions, in environments that require automation and contactless measurement. Keywords: Analog Gauge, Ellipse Fitting, SAM, TrOCR, YOLOv9]]>
Copyrights © 2026
