A Method of Pointer Instrument Reading for Automatic Inspection

In automatic inspection, it is required to automatically read the indicator value of the pointer instrument from images. However, in the cases of different shooting positions and focal lengths, the images will be blurred and deformed, which bring difficulties to indicator value reading. This paper proposes a method for reading the indicator values of pointer instruments automatically. First, we make clear templates for each type of dashboard. Then, we use KAZE to extract the features for the input image, use KNN to match the features from this image and the templates, use RANSAC to reduce the wrong matchings, and get the type and area of instrument based of matching results. Finally, we calculate the indicator value by edge information, pointer line information from probabilistic Hough transform, and labeled prior knowledge. The experimental results show the output values within a degree scale is 96.4% in 560 samples, which meets the requirements of automatic inspection.