Graphics Reference
In-Depth Information
Fig. 12.1
An example of an ASM system for surveillance
Unfortunately, this approach is infeasible because it requires large communication
bandwidths and energy consumption, and long transmission and processing delays.
A feasible approach involves classifiers—localized in the same processing node of
a camera—that are in charge of preprocessing the images. Based on the results
of such preprocessing, the classifiers decide: (1) at which rate to acquire images,
(2) whether or not to discard a specific image, and (3) in case the image is not
discarded, the node to which the image must be sent for further processing and the
resolution at which the image must be transmitted. Then the results of image process-
ing can be exploited to trigger actions that modify the environment under observation
(e.g., some roads are opened or closed) and even the streammining system itself (e.g.,
additional cameras are turned on).
12.3 Challenges in ASM System Design
Key challenges in distributed real-time streammining systems arise from the need to
cope effectively with system overload due to large data volumes and limited system
resources. There is a large computational cost incurred by each classifier (propor-
tional to the data rate) that limits the rate at which the application can handle input
video. Commonly used approaches to dealing with this problem in resource con-
strained stream mining are based on
load-shedding
, where algorithms determine
when, where, what, and how much data to discard given the observed data character-
istics, e.g. burst, desired Quality of Service (QoS) requirements [
4
,
5
,
37
-
41
], data
value or delay constraints [
12
,
15
].
