Water Meter Reading Area Detection Based
on Convolutional Neural Network
Jianlan Wu
(&)
, Yangan Zhang, and Xueguang Yuan
Beijing University of Posts and Telecommunications, No. 10, Xitucheng Road,
Haidian District, Beijing 100876, China
wujianlan@bupt.edu.cn
Abstract. The water meter is a device for measuring the amount of water used
by each household. Remote meter reading is one of the main ways to solve the
waste of human resources caused by regular manual door-to-door access to
mechanical water meter readings. The current use of image acquisition and then
accurate reading of the water meter image is one of the ways of remote meter
reading. In this paper, the convolutional neural network is used to predict the
reading area, and then the non-maximum suppression algorithm (NMS) is used
to remove highly overlapping results from prediction region results to obtain the
position of the reading area. The experimental results show that with using the
method proposed in this paper in the actual application scenario, the IoU of the
images of 1000 test sets are all above 0.8 and then combined with the three-layer
BP neural network for character recognition, the accuracy rate reaches 98.0%.
Keywords: Convolution neural network Á Image detection Á Water meter
number reading
1 Introduction
At present, most water meters are still mechanical water meters, and the update speed
of smart water meters is slow. For users who have not replaced smart water meters, the
work of water meter readings is mainly done by humans every month, which is not
only inefficient, but also has potential safety hazards. A lot of human resources are
wasted, so remote meter reading is required. One of the remote meter reading methods
uses image acquisition and then recognizes the number on the picture.
Image-based remote meter reading is mainly divided into two steps. The first step is
the detection of the reading area and the second step is the recognition of characters.
Among them, the detection of the reading area is the primary premise of character
recognition. Common methods for detecting the reading area are: method based on
Canny edge detection [1–3] and Hough transform [4–9], method based on projection
[10–12].
The method based on Canny edge detection and Hough transform, the Canny edge
detection algorithm is used to obtain the edge of the image. The Hough circle detection
removes the pixel interference between the circular area and the area where the circular
metal casing is located and then finds the water meter character by Hough line
detection. The line of the four sides of the number box is located, and finally, the
© Springer Nature Singapore Pte Ltd. 2020
Q. Liang et al. (Eds.): Artificial Intelligence in China, LNEE 572, pp. 173–179, 2020.
https://doi.org/10.1007/978-981-15-0187-6_20
on Convolutional Neural Network
Jianlan Wu
(&)
, Yangan Zhang, and Xueguang Yuan
Beijing University of Posts and Telecommunications, No. 10, Xitucheng Road,
Haidian District, Beijing 100876, China
wujianlan@bupt.edu.cn
Abstract. The water meter is a device for measuring the amount of water used
by each household. Remote meter reading is one of the main ways to solve the
waste of human resources caused by regular manual door-to-door access to
mechanical water meter readings. The current use of image acquisition and then
accurate reading of the water meter image is one of the ways of remote meter
reading. In this paper, the convolutional neural network is used to predict the
reading area, and then the non-maximum suppression algorithm (NMS) is used
to remove highly overlapping results from prediction region results to obtain the
position of the reading area. The experimental results show that with using the
method proposed in this paper in the actual application scenario, the IoU of the
images of 1000 test sets are all above 0.8 and then combined with the three-layer
BP neural network for character recognition, the accuracy rate reaches 98.0%.
Keywords: Convolution neural network Á Image detection Á Water meter
number reading
1 Introduction
At present, most water meters are still mechanical water meters, and the update speed
of smart water meters is slow. For users who have not replaced smart water meters, the
work of water meter readings is mainly done by humans every month, which is not
only inefficient, but also has potential safety hazards. A lot of human resources are
wasted, so remote meter reading is required. One of the remote meter reading methods
uses image acquisition and then recognizes the number on the picture.
Image-based remote meter reading is mainly divided into two steps. The first step is
the detection of the reading area and the second step is the recognition of characters.
Among them, the detection of the reading area is the primary premise of character
recognition. Common methods for detecting the reading area are: method based on
Canny edge detection [1–3] and Hough transform [4–9], method based on projection
[10–12].
The method based on Canny edge detection and Hough transform, the Canny edge
detection algorithm is used to obtain the edge of the image. The Hough circle detection
removes the pixel interference between the circular area and the area where the circular
metal casing is located and then finds the water meter character by Hough line
detection. The line of the four sides of the number box is located, and finally, the
© Springer Nature Singapore Pte Ltd. 2020
Q. Liang et al. (Eds.): Artificial Intelligence in China, LNEE 572, pp. 173–179, 2020.
https://doi.org/10.1007/978-981-15-0187-6_20