QR-pattern-assisted perspective distortion correction for circular pointer meter image

The perspective distortion will present in captured image when the pointer meter is shot at non-orthogonal angles relative to the meter plane. This geometric deformation significantly affects the accuracy of automatic reading recognition of meter. Therefore, perspective distortion correction for pointer meter image is an important step of intelligent reading recognition. Furthermore, for the circular pointer meters, perspective distortion correction is more difficult. The paper proposes a QR pattern-assisted perspective distortion correction method for circular pointer meter. In this method, the QR pattern placed beneath meter dial is firstly extracted by using the YOLOv4-tiny. Then four corner points of QR pattern are detected through several image processing operations. Finally, based on the relationship of reference points and their corresponding ideal coordinates, the perspective transform is performed for meter image to obtain the distortion-free image. The experimental results of actual meter images demonstrate that the proposed method can achieve good correction results.