Catching Data from Displayers by Machine Vision
Lifeng Yao
1,a
, Jianfei Ouyang
1,b
1
State Key Laboratory of Precision Measuring Technology and Instruments, Tianjin University,
Tianjin 300072, China
a
yaolifeng2006@gmail.com,
b
oyj@tju.edu.cn
Keywords:
Data Catching
,
Machine Vision, Numeric Recognition, Algorithm
Abstract. With the emergence of eHealth, the importance of keeping digital personal health
statistics is quickly rising in demand. Many current health assessment devices output values to the
user without a method of digitally saving the data. This paper presents a method to directly translate
the numeric displays of the devices into digital records using machine vision. A wireless-based
machine vision system is designed to image the display and a tracking algorithm based on SIFT
(Scale Invariant Feature Transform) is developed to recognize the numerals from the captured
images. First, a local camera captures an image of the display and transfers it wirelessly to a remote
computer, which generates the gray-scale and binary figures of the images for further processing.
Next, the computer applies the watershed segmentation algorithm to divide the image into regions
of individual values. Finally, the SIFT features of the segmented images are picked up in sequence
and matched with the SIFT features of the ten standard digits from 0 to 9 one by one to recognize
the digital numbers of the device’s display. The proposed approach can obtain the data directly from
the display quickly and accurately with high environmental tolerance. The numeric recognition
converts with over 99.2% accuracy, and processes an image in less than one second. The proposed
method has been applied in the E-health Station, a physiological parameters measuring system that
integrates a variety of commercial instruments, such as OMRON digital thermometer, oximeter,
sphygmomanometer, glucometer, and fat monitor, to give a more complete physiological health
measurement.
Introduction
With the emergence of eHealth, the importance of keeping digital personal health statistics is
quickly rising in demand. Many current health assessment devices output values to the user without
a method of digitally saving the data. They are including digital thermometer, oximeter, and fat
monitor etc.
In order to output the measurement results in the case of the electronic measuring devices have
no standard output interface, the image processing method based on machine vision is presented,
which can realize the auto-recognition of the numeric displaying on the devices. While acquiring
the images of the display screen, which is often affected by uneven illumination, so that the
acquired images are with gray uniform and mutation. Sometimes, because of the shooting location
could not be kept being vertical with the display screens, but with a certain angle, the images have
different perspectives and tilt angles. That is, affected by the factors like illumination, perspective
and scaling, the acquired images are with non-uniform gray, ranging in size and with a certain
degree of tilt, making it difficult to recognize accurately the numeric displayed on display screens.
There are many methods to recognizing the numeric, such as the pixel-based neural network [1]
and the identification method based on the topological structure of characters, which including the
threading methods. However, SIFT is a matching algorithm based on the scale invariant features of
images. SIFT feature matching algorithm [2] is presented by David Lowe through summing up the
proposed feature detection method [3] based on invariant features in 2004. The algorithm is
extracting the features on DOG scale-spaces and the 2D image space. The features are invariant to
image scaling and rotation, and partially invariant to change in illumination and 3D camera
viewpoint even to affine transformation. The algorithm is well-matching, and the features extracted
by which methods is stable. It can match two images well, while translation, rotation, affine
transformation, perspective transformation, illumination change between them, even the images are
shooting at any angle. In a word, it can match the two images in large differences with features.
Advanced Materials Research Vol. 566 (2012) pp 124-129
© (2012) Trans Tech Publications, Switzerland
doi:10.4028/www.scientific.net/AMR.566.124
All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of TTP,
www.ttp.net. (ID: 202.113.1.151-24/07/12,15:17:42)