Abstract:The requirement of algorithms for prior knowledge is inconvenient. To solve this problem, an autodetection and autorecognition method is proposed, which needs no prior knowledge. To detect the dial area, a rough detection algorithm is designed based on an ellipse detection algorithm and nonmaximum suppression to detect possible areas of the meter. Then, a line segment detector is enhanced by designing the preprocessing and the method of obtaining gradient values and levelline angles. And a centripetal angle constraint is added to the detector when seed points are selected to detect centripetal lines to filter the right area of the meter based on the number of centripetal lines. In the method of recognizing meters, an ellipse fitting method in which points are randomly selected from several sectors is proposed to fit inner ends of centripetal lines to identify scale lines. The segment fusion conditions are set to detect the pointer. The method based on maximally stable external regions is used to detect regions of interesting (ROIs). After identifying ROIs, neighborhood ROIs are connected to combine scale values, which are linked to their nearest main scale lines′ angles. Finally, the reading of the meter is obtained according to the angle of the pointer and the relation between values and angles. Experiments show that the total running time of the algorithm is 063 s, and the probability of error in one scale and in two scales are 80% and 967%, which can meet the actual requirement.