Research on a Visual Reading Method for Pointer Instruments Integrating Geometric Constraints

A pointer meter automatic reading algorithm integrating geometric constraints is proposed. First, the dial region is located by the Hough circle transform, and the dial centre is taken as the pointer rotation centre. Then, segmentation thresholds are selected adaptively according to the pointer colour, pointer pixels are extracted within a restricted annulus, and the pointer direction is determined by angular histogram voting. Next, the recognition result is screened for geometric consistency using the rotation centre, annulus and confidence constraints, and a rejection state is output when the evidence is insufficient. Finally, the meter reading is calculated from the angular proportion of the pointer between the start and end scale marks. A 0 to 6 bar pressure gauge in a public real-image dataset is taken as the object, and the accepted readings are established by two independent reading rounds and a third-round arbitration. The experimental results show that the mean absolute error of 10 test samples is 0.375 bar with a fiducial error of 6.24 percent; for the 8 samples whose pointer direction is correctly judged, the mean absolute error is 0.100 bar with a fiducial error of 1.66 percent; the recognition confidence of yellow pointers is clearly higher than that of dark and light pointers, so the rejection threshold should be set separately for each pointer colour.