Research On Meters Reading Recognition For Patrol Robot

The reading recognition of meters is an effective way to control the production of enterprises and its quality.However,in the process of manual meter reading,not only a large number of human costs are needed,but also the gross and random errors will be caused,which affect the quality of the products and even lead to safety accidents.In this way,companies cannot always get the accurate data.As a result,it is of great significance for companies to research and realize a multi-type meter recognition system which used in patrol robot.The method of reading recognition of pointer-type meters and digital-type meters for patrol robot,based on machine vision,is proposed in this paper.In terms of the reading recognition of pointer-type meters,an improved Wellner adaptive binary method based on local variable threshold is proposed to approximate the uneven illumination interference.Hough gradient method is then adopted here to get a rough meter region segmentation result in the picture.Next,the least-square method is used to fit a circle by the rough meter region abstracted in preprocessed image as well as binary image,and the result of rough segmentation is corrected,according to the fitting result,to get the precise position result of meter.Moreover,randomized Hough transform is used to detect straight lines in the precise position area of binary image and the ranking of these lines is optimized to get the final pointer vector.In addition,the radius reduction method is designed to obtain the scale from the edge of the binary image and calculate the datum scale vector.Finally,the indicating value can be obtained by combining deflection angle and range information,and the deflection angle can be calculated by pointer vector and datum scale vector.What's more,a method for simultaneous recognition of multi-meters' readings with variable viewing angle is also presented in this paper.The improved Freeman coding method is used to encode the edge of the image to get the segmentation result of multiple meters' region,and the reading of these meters are computed by calculation of deflection angle in turn.The experiments verify the effectiveness of this method.As to the reading recognition of digital-type meter,the precise position of digital-type meter can be obtained by using the method proposed in the process of reading recognition of pointer-type meter under uneven illumination.In the next step,randomized Hough transform is used to search the longest line in image edges to correct the skew of digital-type meter image and get a rough segmentation result of digital characters' region.Furthermore,black hat transform is adapted here to highlight the characters in digital region under uneven illumination and to perform binary segmentation by using Otsu algorithm.A block transverse projection method is then proposed to precisely segment the digital character region in binary image and characters are divided by using longitudinal projection in the final precise digital character region.Finally,in order to get rid of the interference such as adhesion in the segmented characters,a seven-segment tube digit recognition method is proposed to classify and recognize the digits.The experimental results show the robustness of this method.At the end of this paper,the requirement analysis and design of reading recognition system for meters are carried out.Aiming at the problem of binary classification,the HOG feature of image is extracted to train SVM model,and the performance of this model is evaluated and analyzed.After obtaining the appropriate model,the HOG features of the meter position area of the image to be recognized are extracted into the pre-trained SVM model for classification.The images to be recognized,according to the results of classification,are processed correspondingly to get the final reading result.Finally,the system is implemented and encapsulated in ROS nodes in order to facilitate the subsequent field testing and application.