Research Of Embedded Meter Reading Recognition System Based On Machine Vision

At present, restricted by working conditions, economics and convenience, in mechanicalmetrology field, there is a heavy use of mechanical and analog instrument, whose reading and datarecording work currently can only be artificial completed. In order to improve work efficiency,reduce the manual intervention and reduce the error probability, instrument visual detection systemwith PC centered is often used. Although they reduce the burden of artificial reading to a certainextent, they still have the disadvantages of large volume, high cost, being not easily and rapidlydeployed and idle computing power. This paper designs and develops a set of embedded meterreading recognition system based on machine vision technology, which can be automatically readand converted existing pointer type and digital water meter dial, to reach automatic identification anddata storage.The hardware system mainly refers to the embedded hardware terminal identification system.The sub system is based on embedded processor, and is equipped with industrial camera with highresolution. The hardware sub system is mainly responsible for the acquisition of image data, therealization of reading recognition algorithm, and providing the necessary man-machine interface etc.Aiming at the support vector machine (SVM)’s disability to have fast recognition speed and thehigh recognition rate at the same time in practical application, this paper puts forward a newidentification algorithm based on image pixel gradient changes. This paper would take the naturalgas meter as the sample table, do positioning, pretreatment, extraction of digital feature pointclassification as well as a series of operations to its dial digital region. Meanwhile, this paperconstructs a classification model based on SVM digital features, uses cross validation idea tooptimize parameters, and at last provides a set of scheme guaranteeing both the recognition rate andidentification speed.Besides, this paper also puts forward an improved pointer type recognition algorithm aimed at a series of problems encountered by the traditional Hough transform processing recognition. Afterextracting the pointer contour, the algorithm positions the dot using the Hough gradient method,positions the direction using the contour method, and calculates the function relationship between thezero line and the instrument indicating value of the pointer and the angle using the Newtoninterpolation method, so as to realize intelligent reading of instrument indicating value. Thealgorithm is a good solution to the interference noise on the image production. The recognition speedis greatly improved while the accuracy is ensured.The software system mainly refers to the server data processing software system. Thissubsystem is built based on.NET Framework4.0, and uses the singleton pattern, observer patternand factory pattern to complete development based on three layer architectures. This system makesgood use of the parallel task provided by.NET Framework4.0(TPL, Task Parallel Library)technology to finish loading, control, management and synchronization of multiple threads.In view of the embedded terminal and specific requirements of background server, this paperdesigns and defines data message used in the process of communication. This paper analyzessecurity of the data transmitted by packet, and then proposes the use of DES symmetric encryptionalgorithm encryption scheme for the transmission of data packets between terminal and server. Thedemand is met while the security of the system is enhanced.Finally, this paper finishes the integration and test of the system. The experimental results showthat the system can well met our demands.