Meter reading recognition method via the pointer region feature

Song Wei,Zhang Wenjie,Zhang Jiaqi,Wang Yuping,Zhou Qing,Shi Weiren

DOI: https://doi.org/10.19650/j.cnki.cjsi.2014.s2.008

2014-01-01

Abstract:Reading recognition algorithms for pointer type meter are vulnerable to illumination, and recognition rate for thicker pointer is not high. In this paper, we propose a reading recognition method based on pointer region. In detail, this approach can be divided into two phases: in the first phase, the proposed method designs Gaussian homomorphic high-pass filter to enhance the adaptability to the illumination change for instrument image, and utilizes image preprocessing methods (including image subtraction, binarization, median filter, image morphology processing et al.) to reduce noise interference, suppress of background region, outstand pointer region. In the second phase, the presented method firstly rotates the meter image, and computes projection of the rotated image on the vertical axis in each rotation angle, then calculates the rotated angle which rotate the pointer of meter image to parallel to the horizontal axis. Whereafter, the deflection angle of pointer can be derived for combining the projection of rotated image on the horizontal axis which the pointer of rotated image parallels to the horizontal axis. Finally, reading recognition is identified by the relation between the pointer angle and the corresponding instruments scale. This approach is simple and stable, the error between the reading of the proposed approach and artificial reading is small, and it can better adapt to illumination changes.