Automatic Correction Method of Industrial Instrument Images Based on YOLOv8 Keypoint Detection and Perspective Transformation

Chuanlei Zhang, Lei Shi, Na Bu, Gongcheng Shi, Weichen Feng, Hui Ma, Zehua Wang

Abstract: In the current process of automatically identifying pointer instruments in substations, it is often encountered that the photographed instruments appear tilted and distorted, making the calibration step essential. This paper presents an automatic correction method for industrial instrument images based on Yolov8 and perspective transformation, aiming to offer a reliable image correction technique for automatic instrument recognition. This approach encompasses three primary contributions: (1) The creation of a Yolov8-pose instrument dataset comprising 524 images. (2) The application of keypoint detection technology into image correction. (3) The proposal of an automatic correction method based on perspective transformation to address square instruments, laying a strong foundation for precise readings. Experimental results demonstrate that the Yolov8-pose model achieves an impressive mAP of 99.5%, and the proposed automatic correction algorithm effectively rectifies the tilt and distortion in instrument images. This method is believed to hold significant potential for enhancing the accuracy of automatic instrument identification, particularly in industrial applications demanding high precision.