Information Technology Reference
In-Depth Information
In the framework, huge amount of RFID and ALPR detection results are
original input of this method. There are four steps in HelpMe framework. Firstly,
TLPs
Extraction,
TLPs
would be extracted from the filtered ALPR detection
results. Secondly, CTPT (Character Transition Probability Table) Generation,
CTPT which generated through the data fusion of RFID and ALPR is used for
predicting potential correct characters. Thirdly,
SLPs
Correction,
SLPs
would
be corrected by Naıve Bayes theorem. Finally,
CLPs
Accuracy Promotion, the
accuracy of
CLPs
would be improved by using RFID detection results through
data fusion. And the optimization on image recognition is not the focus of this
paper.
Detailed
LP
data classification and marking is shown in Fig. 4. Generally,
LPs
are classified into three levels during the process of HelpMe. Original
LP
data is the first level.
TLPs
and
SLPs
which belong to second level are marked
by
TLPs
Extraction. The results of
SLPs
Correction which contains
CLPs
and
EXP
(Exceptions) are regarded as the third level.
EXP
can be categorized to
several types. For example, a vehicle crosses a single camera but stops for a long
time which contradicts with the assumption of floating car. And the trac is so
sparse that little information can be extracted from a camera's neighbors.
Both
TLPs
and
CLPs
would be further corrected by RFID data fusion al-
gorithm during the process of
CLPs
Accuracy Promotion.
Fig. 3.
Overview of the framework of
HelpMe
Fig. 4.
All possible cases of the execution
of HelpMe
A more particular description of the process of HelpMe will be discussed in
the following steps listed below.
TLPs
Extraction.
The process of
TLPs
Extraction aims to identify license
plates that are correctly recognized with high probability. ALPR results are fil-
tered through certain rules to clean the dirty data, such as overexposing pictures
and files corrupted, etc. Distinguishing
TLPs
from big trac data poses a chal-
lenge to real time decision support. Therefore, we reduce the problem complexity
by setting the limit on
k
of
k
-degree neighbors, the number
ʸ
of continuous oc-
currence when
k
is fixed (
ʸ
≤
2
k
+1) and the travel interval
ʔt
among
CA
.