Abstract: Aiming at the difficulty and slow speed of instrument detection and extraction which is caused by factors such as indistinct instrument features and complex background environment, a pointer instrument automatic detection and reading method was proposed based on YOLOv4. Firstly, YOLOv4 was used to detect and extract the pointer meter area, and hough transform was utilized to cut the dial area. Secondly, the dial contour was found in the dial area, and the center of pointer rotation was obtained by fitting the line of the dial contour with the least square method. Finally, the pointer contour was refined and the pointer center line was fitted with the random sample consensus (RANSAC), and the meter reading was obtained according to the rotation angle of the pointer. The experimental results show that the instrument detection accuracy of pointer instrument automatic detection and recognition method reaches 98.6%, the average basic error of reading is 0.72%, and the average recognition time is 80.65ms, which proves the effectiveness and rapidity of the algorithm.

Key words: pointer instrument; YOLOv4;Hough transform; least square method; random sample consensus (RANSAC)