A Value Recognition Algorithm for Pointer Meter Based on Improved Mask-RCNN

The automatic reading recognition technology for pointer meter is widely used in the military, industrial, and aerospace fields. However, the accuracy of recognizing pointer meter is susceptible to some factors during the recognition process, such as uneven illumination, complex background, rotation angle, image blur, shooting angle, scale change and other factors, all of which may result in lower accuracy and usability. In this paper, in order to overcome these difficulties, we make improvements to the existing Mask-RCNN network, and propose a novel deep learning based algorithm using PrRoIPooling instead of RoiAlign to achieve the accurate segmentation to improve the accuracy of meter readings. The main procedure of the algorithm is to first use the Mask-RCNN instance segmentation network to segment the meter dial and the pointer area, and then classify and recognize the meter type while fitting the pointer. Finally, the angle reading method is used to calculate the pointer reading. Experiments show that the algorithm is robust, adaptable and effective to the natural environment.