Research On Reading Recognition For Pointer Meter Based On Machine Vision

Traditional pointer instrument has been widely used in industrial production because of its advantages of simple structure and low production cost.However,the current pointer meter reading mode is mostly manual interpretation,low efficiency.In order to solve the inefficient problem of manual interpretation,automatic instrument reading method has been proposed constantly.This kind of method is based on machine vision,and the efficiency and accuracy are improved compared with manual interpretation.It provides a feasible scheme for the accuracy calibration and regular performance detection of pointer type instrument,and effectively promotes the modernization of the factory.In order to further optimize the reading recognition method of pointer type instrument,this paper proposes an automatic reading scheme of pointer type instrument based on machine vision,including target detection,tilt correction,range recognition,indicator reading,index value detection,and the realization of reading recognition system hardware and software on this basis.The research content of this project is mainly divided into the following parts:(1)Aiming at the problem that pointer instrument target detection method is susceptible to environmental interference,a target detection method based on rotating target detection network is proposed.Based on the horizontal target detection network YOLOv5 s,this method is improved to transform the network into a rotating target detection network which is more suitable for pointer type instrument.Meanwhile,the network performance is improved by Angle regression,small target detection and loss function.Compared with traditional target detection methods and horizontal target detection networks,this method has certain advantages in detection effect and detection speed.(2)Aiming at the problem that the current tilt correction method of pointer type instrument is not strong in generalization ability,the tilt correction method is improved.Make full use of the output information of rotating target detection network to carry out rotation correction and deformation correction for tilted instruments.At the same time,in view of the current reading recognition methods of pointer meters,often need to take the range as a prior knowledge and manually input the problem,a pointer meter range recognition method based on the improved Res Net18 classification network is proposed.Truly achieve the automatic identification of pointer type instrument reading,further reduce the participation of manual.(3)Planning the identification process and analyzing the system architecture according to the actual requirements of the reading recognition system.From two aspects of hardware and software,the corresponding hardware structure and software framework of the system are designed,and the software design of different functional modules is realized.Finally,the reading recognition method of pointer type instrument is applied to the hardware system to realize the picture detection and video detection of pointer type instrument.