Target Re-identification Based on Targtes’
Appearance and Space Features
*
Hua Han, Yuming Wang, Jian Xing
School of Electronic and Electrical Engineering
Shandong Computer Science Center
Shanghai University of Engineering Science
Shandong Provincial Key Laboratory Of Computer Network
Jinan, Shandong Province, China
*
This work is partially supported by the National Natural Science Foundation of China (No. 61272097, 61305014, 61401257), Innovation Program of Shanghai
Municipal Education Commission (No.12ZZ182, 14ZZ156), the Natural Science Foundation of Shanghai, China (No. 13ZR1455200), “Chen Guang” project
supported by Shanghai Municipal Education Commission and Shanghai Education Development Foundation(13CG60), Key Support Project of Shanghai Science
and Technology Committee(13510501400), Funding Scheme for Training Young Teachers in Shanghai Colleges (No. ZZGJD13006), Shandong Province Young
and Middle-Aged Scientists Research Awards Fund, China (No. BS2013DX021); Shandong Academy Young Scientists Fund Project (No. 2013QN037), The
connotative construction projects of Shanghai University of Engineering Science in the 12th Five-Year (nhky-2014-12).
Abstract—Target re-identification of cross-camera is a difficult
problem in the field of multi-camera surveillance, which needs to
be urgently solved. Traditional solutions are mainly depend on
the characteristics of target appearance. But these methods easily
have the problem of low reliability and easily lead to low
matching rate, because of losing targets’ space information. In
this paper, we will use targets’ appearance and space information
to make multi-layer histogram to describe targets, and by
comparing the corresponding histograms to achieve target re-
identification. At last, Experiments of targets re-identification for
different cameras show that the proposed method can achieve
good results.
Keywords-Target re-identification
˗
Multi-layer histogram
Appearance and space feature
I. INTRODUCTION
Target re-identification from cross-camera is an important
research topic in multiple cameras intelligent tracking [1-3].
And the most important hotspot is how to achieve a same
target’s re-identification from different cameras under complex
environment, such as different illumination, different
pedestrian gait, different scales occlusion and so on. As for
target re-identification, it is the question of when target
detections in different views or at different time instants can be
linked to the same individual [4].
Nowadays, multiple scale camera network is applied to
tracking targets over a complex area. Various types of camera
overlap and non-overlap can be employed in multi-target
tracking. In this context, tracking means that computers or
cameras can be able to point out the same target from one
camera to another when the target disappears from the former
camera and appears in the latter one. The ability to achieve
quickly identifying in complex environment depends on
whether we can find the invariant in time and space domains.
One common processing approach is extracting targets’
features to achieve re-identification.
Now, the correct rate of target re-identification using the
most advanced algorithms based on VIPeR (Viewpoint
Invariant Pedestrian Recognition) database is still below 20%,
which is at a relatively low level [5]. This makes target re-
identification methods become more dependent on features of
targets’ external appearance, in the situation of low reliable
spatial-temporal information.
From quantity literatures, we can know that the existing
works predominantly focus on feature extraction and
representation. An appearance-based target reacquisition
approach relies on estimating a signature representing the
identity of the target being tracked. Histogram-based
descriptors are ubiquitous tools in feature description. The
histograms are a widely used appearance describers, but the
main drawback is that they lose spatial information of the color
distribution which is essential to discriminate different moving
targets. For example, histogram-based methods can not tell a
person wearing a red shirt and green pants from another person
who dresses in green shirt and red pants. So, targets’ space
2015 Fifth International Conference on Instrumentation & Measurement, Computer, Communication
978-1-4673-7723-2/15 $31.00 © 2015 IEEE
DOI 10.1109/IMCCC.2015.369
1738