Graphics Reference
In-Depth Information
Fig. 12.4 Illustration of object or face recognition via adaptive wireless video transport to a remote
computing server
wireless cluster . A server running openstack or Hadoop (or a similar runtime envi-
ronment suitable for cloud computing) [ 25 ] is used for analyzing visual data from
numerous wireless clusters, as well as other computing tasks unrelated to object or
face recognition.
Each device can adapt the encoding bitrate, as well as the number of frames
to produce (with the ensemble of N such settings comprising the set
A =
{
), in order to alleviate the impact of contention in the WLAN. At
the same time, the visual analysis performed in the cloud can be adapted to scale
the required processing time to alleviate the impact of task scheduling congestion
in the cloud [ 25 , 30 ], with the sets of contention and congestion levels represented
by the discrete sets
a 1 ,
a 2 ,...,
a N }
, respectively. In return, each device receives from the
cloud a label that describes the recognized object or face (e.g., the object or person's
name), or simply a message that the object or person could not be recognized. In
addition, each device or wireless cluster can also receive feedback on the experi-
enced WLAN medium access control (MAC) layer contention and the cloud task
scheduling congestion conditions.
Thus, the “reward” for each device is the recognition result at each time step.
Given that each wireless access point and the cloud computing infrastructure serve
many more requests than the ones from a given cluster of devices (as illustrated in
Fig. 12.4 ), we can safely assume that for each device, the wireless contention and
T
and
G
Search WWH ::




Custom Search