Research on the Visual Recognition Method of Pointer Water Meter Reading

For a long time, water companies have adopted manual meter reading methods when they settle water bills. Manual meter reading is inefficient, and there are often large subjective errors in it. Therefore, that water companies identify water meter readings automatically by computer technology and image processing technology is the future trend. This paper designed a kind of method to identify the water meter reading accurately. The method preprocessed collected water meter dial images by two-dimensional Gaussian filtering firstly. Then the pointer areas were separated from them by threshold and area growth method based on the difference of gray value. Finally, using Bresenham algorithm to calculate deflection angle between pointer center line and the zero scale which realizes the recognition of the reading. To enhance the algorithm's robustness to lighting, this method added the dynamic threshold mechanism and the pixel repair mechanism in the recognition process. This method was deployed on the Raspberry Pi 4B development board equipped with Raspbian GNU/Linux 10 "Buster" operating system, and the relationship between image resolution and reading accuracy was discussed.