Reporting Road Problems in Smart Cities Using OpenIoT Framework - Interoperability and Open-Source Solutions for the Internet of Things - page 172

misleading data, can be blocked. On the other hand users who provide reliable and

relevant reports get credibility points, which can be later transformed into some
incentives.

-

Feedback. Users who send data want to understand what happens next. If they
don

￿

t receive an acknowledgement that their request was processed and their work
is needed, they will not send reports any more. Citywatcher Android application has
a form with reports on all shared videos, so the user can check if his/her contri-
bution was used.

'

Designing, running and automating a processing center is also a big challenge. First off
all we need some speech recognition mechanisms, so the system could get some
semantics without human intervention. Speech recognition can be done both on the
side of pervasive cameras and on the middleware side. We choose to make it on the
middleware side, as it would be easier to maintain, update and tune. Besides, such
mechanisms could be energy consuming and it is better not to produce extra load on
user

s devices. After speech recognition the system can try to apply semantic reasoned
and make some decisions. These decisions include answering questions like:

'

￿ Is this report unique or is it redundant?

￿ How urgent is the problem?

What district and what service are responsible for solving the problem?
At rst, report processing can be done manually by experts. Later, when the business
logic becomes computerized, many techniques can be used for process automation and
reduction of the human involvement. Context reasoning techniques, which will be used
in future releases of CityWatcher, can be broadly classi

￿

ed into six categories: (a)
supervised learning, (b) unsupervised learning, (c) rules, (d) fuzzy logic, (e) ontological
reasoning and (f) probabilistic reasoning. For example, supervised learning methods
include arti cial neural networks, Bayesian networks, case-based reasoning, decision

tree learning and support vector machines [ 22 ]. Such methods must be supported by the
middleware platform. We will discuss the choice and features of the middleware
platform in the next section. The speech recognition engine is an important point for
automatic report processing. The best way for getting results without having serious
troubles with tuning and administering the engine is using a cloud service like Yandex.
SpeechKit [ 23 ].

4 CityWatcher Architecture and Implementation

The proposed system is divided into three main parts: ICOs, middleware and client
software as shown in Fig. 3 . ICOs include the smartphones, VMSCs, smart cameras,
cars video registers, etc. All participating ICOs record video via a special application.
The prototype is an Android app. Application features used in the discussed sce-
nario include Video recording, Video annotating (GPS coordinates, time), auto register
with middleware, constructing reports and sending them to the cloud, viewing the
information about reports that were already sent. Screenshots of CityWatcher appli-
cation for Android are presented in Figs. 4 and 5 . It is a working prototype app.