Bidirectional Matching Algorithm for Target Tracking Based on SIFT - Advanced Research on Computer Education, Simulation and Modeling - page 253

Bidirectional Matching Algorithm for Target Tracking
Based on SIFT
Zhenxing Wu, Jingling Wang, Chuanzhen Li, Yue Yan, and Chen Chu
Information Engineering School, Communication University of China,
Beijing, 100024, China
{Wzx,wjl,lichuanzhen}@cuc.edu.cn, [e-mail]
Abstract. The SIFT is mainly used to detect the feature points of an image, and
achieve a match between two images. Because of its stable ability in matching, it
often gets an excellent performance on images match. This paper is mainly to
study the performance of the SIFT on moving target tracking, we propose a
bidirectional algorithm based on SIFT, and make it automate select the best
template when there are many templates in the data base. Our results of the
experiments show that the bidirectional matching algorithm we proposed here is
more accuracy and stable than the previous unidirectional matching algorithm.
Keywords: SIFT, Scale-space, feature detecting, bidirectional matching,
template selecting.
1 Introduction
Because of the stable matching performance of SIFT, and it can keep invariant to the
change of the light and the scale, it is widely used in Face Recognition, Images Match
and Synthesis. But it is rarely used in target tracking, this is mainly because its match is
usually dependent on its template. Most of the time, however, it is very hard to achieve
much when the target deformation happens [1]. This article is mainly to study the
performance of the SIFT on moving target tracking.
In the article, we propose a kind of bidirectional algorithm under SIFT features, by
doing this, we can avoid the situation when two or more points in one image both match
with one point that in the other image, and the quantity of the right matched points
are not decrease too much. At the same time, because the SIFT features can keep
invariant to the change of the light and the scale, and it not need to know too much
about the target that you want to track, it will have a good performance in target
tracking .
This paper is organized as following: In section 2, we briefly describe how to extract
SIFT features from an image. In section 3, we propose the bidirectional matching
algorithm. Section 4 is the results of the experiment and the analysis. Section 5 is the
conclusion of the paper.