[Studies in Computational Intelligence 481] Artur Babiarz, Robert Bieda, Karol Jędrasiak, Aleksander Nawrat (auth.), Aleksander Nawrat, Zygmunt Kuś (eds.) - Vision Based Systemsfor UAV Applications (2013, Sprin

206 Practical Applications of Classification Algorithms Used in Video Streams
Main aim of design and construction of UAVs is their practical application. A
matter of great importance for real life solutions is utilization of them as a part of
complex high level systems. An essential feature of such systems is a great deal of
automation achieved often by using classification algorithms. They provide the
basis for such advanced systems. However, their design is a lengthy process and
requires intensive evaluation using both simulation and experimental data. Even
then, it is often necessary to apply amendments based on the actual real life experience
of the application.
The first stage of testing such systems is utilization of data acquired using advanced
simulation environments. Characteristic feature of recent simulators are
their advanced physics engines. It is possible to implement an UAV navigation
algorithm following established trajectory and at the same time observe video
streams acquired from virtual cameras mounted under UAV models implemented
in the simulation environment. Obtained positive results using simulation is often
considered as a first step of development detection and classification algorithms.
Misrecognition can sometimes lead to irreversible consequences therefore in
order to improve the quality of recognition frequently fusion of data from multiple
sources of information is used. Unfortunately increased amount of information
often becomes impossible for real time processing using current hardware. It is
therefore necessary to use dimensional reduction and clustering algorithms in
order to reduce the amount of processed data.
An example of using aerial vision from unmanned aerial vehicles is recognition
of people based on their gait features. It is a complex problem that might be solved
using already hinted dimensionality reduction algorithms both in linear version
like PCA, as well as non-linear, such as MPCA.
However it is worth mentioning that before the video stream from a camera
mounted on an UAV it is necessary to wirelessly transmit to human operator using
ground base station. This process has to be reliable in order to allow UAV practical
application. At the same time due to safety considerations it has to be encrypted
because of due to the high risk of wireless data capture. In addition, video data
transmission requires a significant network bandwidth. Communication over long
distances is charged a considerable limitation in this regard. Therefore it is necessary
to use a dedicated image and sound compression algorithms.
Another challenge is to design and implement dedicated communication devices.
They allow autonomous communication between the miniature unmanned
vehicles in a group without the need for continuous ground control. The chapter
presents such device. There is described physical layer and software layer as well
as tests with a group of UAV’s.
All of the above issues have been included in the chapter. In addition, for each
issue a current challenges are presented. Finally, solutions developed by research
teams are also presented. Results, thoroughly discussed examples and valuable
advices might be found within them.