609 
A NEW FRAMEWORK OF MOVING TARGET DETECTION AND TRACKING FOR 
UAV VIDEO APPLICATION 
Wenshuai Yu a ’*, Xuchu Yu b , Pengqiang Zhang, Jun Zhou 
a Institute of Surveying and Mapping,450052, Zhengzhou, Henan, China-ws_yu@yahoo.cn 
b Institute of Surveying and Mapping,450052, Zhengzhou, Henan, China -xc_yu@yahoo.com 
WGS, WG III/5 
KEY WORDS: Image Processing, Computer Vision, Motion Compensation, Motion Detection, Object Tracking, Process Modeling, 
UAV Video 
ABSTRACT: 
Unmanned aerial vehicle is a new platform for remote sensing, and the primary sensor of it is video camera. Video, also could be 
called dynamic image is the most important data format which obtained by unmanned aerial vehicle. The combination of video data 
and UAV provides a novel remote sensing pattern. Moving target detection and tracking is an important technique of video 
processing for its huge potential in military and other applications. The technique always contains three basic parts: motion 
compensation, motion detection and object tracking. Each part adopts kinds of technical methods to solve the problems in respective 
fields. The paper, based on the analysis of the algorithms related to the technology, presents a new framework of it. Different from 
other moving target detection and tracking frameworks, the framework performs a parallel processing among the three sections by 
including collaboration control and data capture modules. Comparing with other frameworks, it is more suitable to the UAV 
applications, because of its advantages such as transferring parameters instead of real data and offering interface to user or exterior 
system. 
1. INTRODUCTION 
Unmanned Aerial Vehicle (UAV) is a new developing remote 
sensing platform, and different from other platforms, for 
example satellite or airplane, it carries video sensors. So video 
data is the main information got by UAV. Video could be 
interpreted as dynamic image, and dissimilar to static image, it 
can reflect motion information through the changing of gray- 
level. An important research field of video processing for UAV 
application is moving target detection and tracking. In actual 
environment, the moving targets could be vehicles, people or 
aircrafts, and in some special conditions, these targets might be 
interesting and valuable. But the problem that detecting the 
targets from the complicated background and tracking them 
successively is a tough work. 
There many technique methods on moving target detection and 
tracking. Most of them analysed the problem under the 
condition of static background, for the stillness of background 
makes the detection and tracking comparatively easier, and 
these kinds of method can be used in some applications such as 
safety monitoring. Contrasting to them, it is much more difficult 
for target detection and tracking with moving background. 
Especially for UAV video data whose background changing 
rapidly and always has complex texture characteristic, it is 
really a challenging task to solve the technical problem. 
For moving target detection and tracking using UAV video, a 
rather reasonable technical approach is adopted widely. Firstly, 
in order to compensate the background motion caused by 
movement of camera, stabilizing the background through the 
frame-to-frame registration of video image sequence would be 
taken as a precondition of detection and tracking. A significant 
product the panoramic image is built in the same process. 
Secondly, basing on the stabilization of background and 
employing proper methods, the next operation is separating the 
target image from the background to realize detection of 
moving target. Finally, moving target tracking is locating the 
object in image by means of modeling the target according to 
target’s feature property and choosing appropriate tracking 
method. 
According to the technical approach mentioned above, the 
technique can be divided into three sections: motion 
compensation, motion detection and object tracking. It always 
takes the three parts as a serial course and implements them one 
after another in a processing. Actually, for there are mutual 
activities between different sections of the technique, it is not 
necessary to process the technology orderly, which means 
executing it step by step. So it not only needs a framework to 
integrate all these parts, but also requires the framework more 
effective and practical. 
2. MOTION COMPENSATION 
Motion compensation is the basic part of the technique, 
especially for moving background video. It estimates the ego- 
motion of camera and compensates the background motion of 
image, and through this way, it makes the moving objects more 
obvious and the detection of target easier. There are two kinds 
of approaches adopted, one is feature-based methods, and the 
other is flow-based methods. Though the latter one has rigorous 
theory foundation, the former one is more popular. Feature- 
based methods extract features and match them between image 
frames to fit the global motion model of video image sequence. 
* Corresponding author. Tel.:+86-13526657654; E-mail address:.ws_yu@yahoo.cn.