In this area of research, the automated recording of meter readings by mechanical meters is examined and several processes for the autonomous recognition of the meter reading developed. The process developed here uses the various reflexions or grey values of the different numbers to recognize the meter reading. These are determined through LED lighting and measurement with phototransistors. The signals are read in without further strengthening with the internal analog digital converter of a microcontroller. Through different placing of the optoelectronic components, a multi-dimensional measurement vector is formed which is classified through a process which has been developed. In this way various classification algorithms, such as, for example, the Bayes classifier and the SVM, are examined. In the research project, both the problem of a direct number recognition without additional teaching, as well as the problem of number recognition with training values available, which however requires movement of the number cogwheels to be classified, are examined and solved.
The graphic shows the statistical distribution of two characteristics for the classification. Through the mechanical structure of the meter, the cogwheel has one or two stable states. This can be seen clearly in the distribution of the characteristics.