|
Abstract
|
|
|---|---|
| This paper presents a novel robust visual tracking framework, based on discrim- inative methods, for Unmanned Aerial Vehicle (UAV) to track an arbitrary 2D/3D target with no prior knowledge at real-time frame rates, that is called the Adaptive Multi-Classi?er Multi-Resolution (AMCMR) framework. In this framework, adaptive Multiple Classifers (MC) are applied to classify the compressed positive and negative samples in downsampling-based Multiple Resolutions (MR) structure in order to update and detect the current target. And the sample importance has been integrated into this framework to improve the tracking stablity and accuracy. The performance of this framework was evaluated with the Ground Truth (GT) in di erent type of public image databases and real ight-based aerial image datasets ?rstly, then the framework has been applied in the O shore Floating Platform (OFP) Inspection task. The evaluation and application results show that this framework is more robust, e?cient and accurated against the existing state-of-art trackers, overcoming the problems generated by the challenging situations such as obvious appearance change, variant illumination, partial/full target occlusion, blur motion, rapid pose variation and onboard mechanical vibration, among others. To our best knowledge, this is the ?rst work to present this framework for solving the online learning and tracking freewill 2D/3D targets problems in the UAV. | |
|
International
|
Si |
|
Congress
|
2nd IFAC Workshop on Research, Education and Development of Unmanned Aerial Systems (2013) |
|
|
960 |
|
Place
|
University of Technology of Compiegne, Compiegne, France |
|
Reviewers
|
Si |
|
ISBN/ISSN
|
1474-6670 |
|
|
|
|
Start Date
|
20/11/2013 |
|
End Date
|
22/11/2013 |
|
From page
|
99 |
|
To page
|
106 |
|
|
Research, Education and Development of Unmanned Aerial Systems |