In an article published in Sensors, researchers demonstrated a systematized scale-mark-based gauge reading (SGR) technique to achieve automatic gauge reading of values from various gauges.
The proposed gauge reading technique employed principal components analysis (PCA) to identify the primary eigenvector of every scale mark. Gauge center detection of the scale marks was achieved by recovering the gauge center from the intersection of these eigenvectors. The corresponding angles of the gauge center were determined by locating the gauge pointer's terminus.
The associated dial values were recovered to align with the scale markers using optical character recognition (OCR). Finally, the linear interpolation of the angles was applied to generate the gauge reading value.
Four videos of realistic environments with perspective distortions and light were used in the investigations. YOLOv4 initially detected the gauges in the video. The discovered portions were clipped to create the input images.
The findings demonstrated that the novel gauge reading technique could successfully achieve automatic gauge reading of the values. Scale-mark-based gauge reading typically had errors of less than 0.1% in a typical environment. In contrast, the typical inaccuracy of scale-mark-based gauge reading was less than 0.5%, even in environments where perspective distortions and light were abnormal.
Automatic Gauge Reading and its Contributions to Industrial Automation
The tremendous positive effect of industrial automation on industrial evolution has made it an unstoppable trend in the modern setting. However, several industries still rely on numerous outdated analog gauges. Manual gauge reading requires traveling to the gauges and is time-consuming, costly, labor-intensive, and prone to frequent reading and recording errors.
Automatic gauge reading is a crucial research area. It takes pictures of the analog gauge with a camera and then applies computer vision techniques to read the gauge values effectively.
Although the bezels on analog gauges can have a variety of designs, there are typically two types of gauges in their charts, arc and circular. Moreover, every gauge has five characteristic features, including a bezel, a pointer, a scale mark, a gauge center, and a dial value.
In this study, the bezel's shape was insignificant and not considered. The gauge center was often placed in the center of a circular gauge. On the other hand, an arc gauge contained it in the bottom right corner.
Circular and arc gauges typically had unique gauge reading procedures, but these procedures were not sufficient for use with other types of gauges. All gauge types might be suitable for deep learning neural networks, but not all gauge types were made for simple training deep learning neural networks.
The systematized approach presented in this research could be used to automatically read the values from an analog circle and arc gauges. This scale-mark-based gauge reading utilized every scale mark as a connected component (CC) for gauge center detection. The area, ratio of CCs, and compactness were the three properties in CCs that were utilized to extract the scale markings. Moreover, the eigenvectors corresponding to the scale markings were recovered using principal components analysis.
Scale-mark-based gauge reading was primarily used in the spatial domain. The characteristics of the angle and area of scale marks were used to determine the primary scale marks. An open optical character recognition (OCR) system called Keras optical character recognition extracted the dial values.
Keras optical character recognition inadvertently determined the dial readings with a negative or floating number because it is not intended mainly for the values on the gauges. The error values of Keras optical character recognition were automatically corrected by scale-mark-based gauge reading using the common difference feature.
The schematic for the scale-mark-based gauge reading architecture was split into two halves.
Gauge detection formed the first section. The gauge reading technique in the second section consisted of four stages- (i) preprocessing, (ii) gauge center detection, (iii) binding (between dial readings and scale marks), and (iv) angle-based interpolation.
Scale-mark-based gauge reading consisted of cameras placed perpendicularly to the gauges' fronts which continued reading the gauge values. With YOLOv4, two or more perpendicularly oriented images were supplied to the system for gauge center detection.
Scale-mark-based gauge reading could be used with arc and circle gauges. The suggested algorithm used various characteristics of a gauge to identify the CCs of the scale marks. The principal eigenvectors of the scale markings intersected at the gauge center, a property used by the gauge reading technique for the gauge center detection. Using principal components analysis, the eigenvectors of the CCs were discovered. The gauge center detection was then obtained by intersecting these primary eigenvectors.
The gauge reading technique treated each minor scale mark as a CC. Principal components analysis was used to extract the appropriate eigenvectors of minor scale markers.
Keras optical character recognition was a simple technique for dial value detection. The recognition of some scale marks was hindered by Keras optical character recognition. This issue was solved by utilizing a method to get rid of scale marks before using Keras optical character recognition. In this method, the gauge center detection was used to extract a circle with a radius of d2 so that Keras optical character recognition could identify the dial values.
The experiments in this paper used four test videos. Videos one, two, and three were obtained from the internet and included circular gauges, whereas video four was created by the authors and featured an arc gauge.
Three clip shots from each of the four videos served as the test data. Manual marking was used for the ground truth of the gauge center detection. The gauge reading algorithm's error was calculated. According to the experimental findings, the typical reading-value error was 0.79%. SGR, however, performed the best with an inaccuracy of 0.13%.
Prospects of the Novel Automatic Gauge Reading System
In this paper, a generic automatic gauge reading algorithm was primarily proposed. This algorithm could use principal components analysis to determine the central position of arc and circular gauges.
Consequently, the primary dial values and scale marks were retrieved and bound together. The proposed gauge reading technique also solved the concern regarding dial values that were a negative or floating point.
The results of the experiments showed that the suggested algorithm performed automatic gauge reading well for both circular and arc gauges. As a result, the scale-mark-based gauge reading further automated the system since essential parameters were not manually entered.
In the future, the authors intend to research various gauge types to uncover more details, such as the measuring units that appeared in the gauges and confirm the suggested methodology's effectiveness.
Wang, C. H., Huang, K. K., Chang, R. I., Huang, C. K. (2022) Scale-Mark-Based Gauge Reading for Gauge Sensors in Real Environments with Light and Perspective Distortions. Sensors, 22(19), 7490. https://www.mdpi.com/1424-8220/22/19/7490