Information Technology Reference
In-Depth Information
3N16T
0.0626
0.0637
0.0638
0.0672
0.0674
0.0692
0.0719
0.0848
0.0937
0.1014
0.1054
0.1156
0.1283
0.1531
0.2146
0.2164
0.2250
0.2267
0.2355
0.2422
0.2616
0.00
Master/Slaves
3N12T
Unbalanced
2N8T
Taskmasters
Unbalanced
2N20T
Taskmasters
Balanced
2N8T
3N20T
Without
Parallelization
3N24T
1N4T
2N20T
3N20T
3N24T
3N16T
3N12T
2N8T
3N20T
3N16T
2N20T
3N12T
3N24T
Sequential
1N1T
0.05
0.10
0.15
0.20
0.25
0.30
Time [s]
Fig. 9. Execution times to process two consecutive frames of each implemented con-
figuration using the data set composed of own images
5 Conclusions and Future Work
In this work, a parallel implementation of an algorithm able to estimate the
orientation of a unmanned aerial vehicle was presented. The estimation is per-
formed by using a remote processing of images taken from an on-board camera.
A multilevel decomposition method to process the images in an heterogeneous
set of desktop computers was proposed. This method uses an alternative hier-
archy called Master/Taskmaster/Slaves which has as main goal the reduction of
the messages sent over Ethernet in order to minimize the latency of the network.
To test the algorithm two data sets were used.
The results show that using the proposed hierarchy and the multilevel data
decomposition method the speed of the process using the traditional hierarchy
Master/Slaves can be doubled. This method can be used for any type of parallel
implementation with large amounts of information passing from one node to
another. Moreover, for the evaluated case of an UAV along with a ground station,
given the processing speed achieved, a close loop for full control of the yaw angle
oftheUAVcanbeimplemented.
Currently, as a direct application of the research reported in this work, a full
orientation control loop for the QA3 quadrotor is under development.
 
Search WWH ::




Custom Search