2017 IEEE 13th International Conference on Electronic Measurement & Instruments ICEMI-2017
Research on Adaptive SIFT Algorithm in Image Matching of Pipeline Inner
Surface Visual Measurement
Li Zhonghu, Zhang Lin, Wang luling,Yan Junhong,Wang Jinming
School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou 014010, China
Email:lizhonghu@imust.edu.cn
Abstract –SIFT algorithm has many advantages such as high
matching rate and good robustness. However, it still has same
imperfect. Due to the large number of key points and large
amount of computation, the processing speed is very slow. This
article propose an improved SIFT matching algorithm with
threshold and adaptive. In this algorithm, the threshold size is
changeable. The number of feature points and the spatial
domain are also controllable. The matching accuracy and
speed can be improved by optimizing parameters. Then stereo
image matching of inner surface of pipeline can be realized.
Experimental results show that the matching rate and the
matching speed are improved obviously. The algorithm can be
applied to the matching of the inner surface corrosion images.
Keywords –Corrosion Detection; Machine Vision; Image
Processing; Stereo Matching; SIFT Matching
I. INTRODUCTION
Pipeline transportation is a new and economic mode
of transportation and plays an important role in the
national economy. With the rapid increase of the number
of pipe storage and the gradual extension of the using
time in our country, the safe operation of pipeline
transportation system has attracted extensive attention in
the industry. Pipeline leakage caused by pipeline
corrosion has become the most frequent accident in the
pipeline industry, which will not only result in a large
amount of waste of resources and serious environmental
pollution, but also may cause significant casualties and
property losses. Therefore, regular inspection of defects
such as pipeline corrosion is of great significance for
reducing losses and ensuring the normal operation of
pipeline.
In recent years, with the development of machine
vision, it is possible to make 3D inspection and 3D
reconstruction of pipe inner surface defects, which
provides new ideas and methods for the detection of inner
surface defects in pipes
[1]
. In the machine vision system,
two different locations of the camera or transforming the
same camera position to shoot space objects get two
different objects in different angles and location of the
two-dimensional images. The pixels from the same
spatial object point are matched in the two images, and
the three-dimensional coordinates of the spatial point are
obtained by matching the parallax of the pixel. Stereo
matching is the most important for 3D reconstruction
based on binocular vision, and it is also the difficulty of
system realization. The success of stereo matching
directly affects the final result of 3D reconstruction, and
it is very important to select the appropriate stereo
matching algorithm. Aiming at this problem, this paper
proposes an adaptive SIFT matching algorithm to achieve
the correct matching of the stereo image pairs on the
inner surface of the pipeline.
II. SIFT FEATURE MATCHING
ALGORITHM
SIFT(The Scale Invariant Feature Transform), the
scale invariant feature transform, was proposed by David.
Lowe in 2004. It is a method that can detect the local
feature of an image effectively. It is invariant to image
translation, rotation, scaling and even affine
transformation. And it is also a very classical matching
algorithm
[2-4]
.
A. The formation of DoG Pyramid, the construction
scale space
DoG Pyramid, also known as the Gauss difference
Pyramid, is derived from the scale transformation of the
original image, thus the representation sequence of the
image scale space can be obtained:
(1)
( )/ 2
2
1
( , , )
2
x y
G x y e
(2)
In the formula, I (x, y) represents the pixel coordinate
value of the image; G (x, y, sigma) is the Gauss kernel
function.
The Gauss differential kernel at different scales is
made convolution with the image I (x, y), and the concept
of DOG Pyramid is introduced, the scale space function
can be expressed as:
( , , ) ( ( , , ) ( , , )) ( , )
( , , ) ( , , )
D x y G x y k G x y I x y
L x y k L x y
(3)
The principal contours of scale space are extracted as
feature vectors, which can be used to extract image
features such as edges and corners.
B. Space extreme point detection
Comparison of each pixel and its adjacent points (2D
image space is 8 points in 3 * 3 neighborhood, and the
scale space of the same group is 2 x 9 points of two
images adjacent to the upper and lower levels) image
domain as well as the size of the scale space. The purpose
978-1-5090-5035-2/17/$31.00
c 2017 IEEE