Research On Automatic Reading Methods Of Industrial Field Pointer Meters Based On A Patrol Robot

With the progress of society and the development of science and technology,people’s demand for automation and intellectualization of industrial production is gradually increasing,and more and more robots are being used in industrial fields.Pointer meter is still widely used in industrial field because of its simple structure,low cost,high precision and strong anti-interference ability.However,there are some problems in manual meter reading,such as inefficiency,difficulty in electronic data storage,false detection and missed detection caused by personnel fatigue,and it is difficult to read with a small number of fixed cameras because of the wide distribution of pointer meters in industrial fields.Therefore,it is necessary to use a mobile patrol robot for automatic reading.In view of the problems of difficulty in extracting meter region and pointer center line caused by interference of complex background environments and angle reading error caused by tilt photography,deep learning combined with traditional computer vision method is proposed in this thesis.The proposed automatic reading method of the pointer type meter is implemented on a patrol robot platform.The main work is summarized as follows:(1)A meter automatic reading process combining deep learning with traditional computer vision is proposed,and an image data set for pointer meter is constructed.By extracting key information and removing interference information,the meter reading task can be simplified and the calculation amount can be optimized.Using the image dataset for pointer meters,a YOLO V3 object detection model for pointer meters is trained.The YOLO V3 object detection model based on deep learning is used to extract the meter from complex backgrounds.(2)A pointer region and a pointer center line extraction method is proposed.The method includes an improved image preprocessing algorithm,a probability-circle pointer region extraction algorithm,an iterative query domain selection algorithm,and a pointer center line extraction algorithm.It can effectively solve the problem of center line extraction failure caused by image blurring,tilt angle shooting and uneven illumination in long-distance shooting.(3)A meter reading correction method is presented.The method obtains the spatial transformation matrix through the plane construction of AprilTag,then corrects the key points by perspective transformation through the transformation matrix.An algorithm is proposed to build the virtual projection coordinate system of probability-circle center,and the meter reading by the angle method is calculated.(4)The feasibility of the proposed method is verified by real-robot experiments.A model of pipeline devices in natural gas stations is made as a simulated scene in industrial fields,and the method proposed in this thesis is verified by a series of experiments in the simulated scene by a self-made mobile patrol robot.Experimental results show that,the proposed method has relatively strong accuracy,stability and real-time performance,and it could be applied to industrial fields to complete the automatic reading task of pointer type meters.