| Traditional meters are widely used in production life,and the automatic reading of traditional meter readings is of great significance for meter inspection and industrial process monitoring.Compared with manual reading,the use of machine vision technology for reading recognition is not only efficient,low cost and data visualization,but also provides a practical solution for modernization of traditional factories,inspection of meter measurement accuracy and spot check of long-term operation instrument performance.In order to realize the automatic reading of the traditional meter readings,this paper proposes the automatic recognition of the meter readings based on machine vision technology,and researches the corresponding denoising methods for the main noise suffered by the meter images in the industrial environment,as well as the automatic detection of the meter dial area,the automatic recognition of the pointer and digital meter readings,and the software and hardware implementation of the automatic meter reading system.The main research content of the subject can be divided into the following parts.(1)An improved binarization method based on the quadratic moving average method for estimating background illumination is proposed for the threshold segmentation problem of inhomogeneous illumination images.The method uses the quadratic moving average prediction method to predict the trend of grayscale sequences in one-dimensional image space,finds the dividing point between foreground and background according to the prediction trend for background estimation,then separates the target image using the background difference method,and finally uses the maximum inter-class variance method for global segmentation to obtain segmentation results.After testing and analysis,the algorithm achieves an effective balance between segmentation effect and segmentation speed compared with other threshold segmentation methods.(2)To study the identification and localization of meter dial position,we propose a dial display region finding algorithm based on the meter contour by using the internal information of the meter contour and the nested relationship between the contours.It is experimentally verified that the algorithm is not affected by the shape,position and size of the meter.The corresponding algorithm is proposed for digital meter display digit recognition,and the character recognition is effectively realized by binarizing the meter image,locating the digital region,tilt correction,character segmentation,rough matching,and exact matching steps.The recognition method is experimentally verified to be less computationally intensive and with higher recognition rate.An automatic recognition algorithm based on scale profile fitting is proposed for the recognition of traditional pointer-type meters.The algorithm achieves the reading recognition of pointer-type meters through the steps of meter image preprocessing,meter ellipse fitting,pointer position detection,scale line position sorting and meter index calculation.The algorithm is experimentally verified to have strong anti-interference ability,small reading error and meet the real-time requirements.(3)According to the actual requirements of the reading recognition system,the overall architecture and recognition process of the system are analyzed,in which the requirements of the meter reading recognition module are focused.Then the corresponding hardware structure and software framework are designed according to the requirements of the meter reading recognition module,and the function design is carried out for the software framework and the software design of the module is realized through Qt Creator and Open CV programming.Finally,the improved automatic recognition algorithm is applied to the recognition module and the corresponding test experiments are made to realize the automatic detection of meter readings and visualize the whole detection process. |
