Novel Robust Perspective Rectification for Automatic Circular-Meter Reading

As the camera lens plane and the meter plane are often not in parallel in reality, the captured images by the camera would have distortions to negative impact on automatic meter reading. In this work, a novel robust automatic contour-based perspective rectification scheme is proposed to enhance automatic reading of circular-pointer meters. Our new scheme can be divided into two major steps, namely, contour rectification matrix estimation and perspective correction of region of interest (ROI). First, the deformed circular-meter contour can be delineated by edge detection in conjunction with curve-fitting to remove the peripheral noise/outliers. Then, a contour perspective rectification matrix can be established upon the aforementioned fit contour. The ROI of the focused circular meter is finally restored using such a contour perspective rectification matrix. The theoretical analysis and experimental results show that our proposed novel robust perspective rectification approach can lead to a higher accuracy than the conventional four-point projective transformation method and spatial transformer network (STN) in terms of rectification error and reading accuracy. The average reading accuracy resulting from our proposed new scheme over hundreds of photographs can reach up to 98.53%.