Information Technology Reference
InDepth Information
Suppose the number of real occurrence of
LP
appeared in its 0,1,2,...,
k
degree
neighbors is
ʸ
,when
ʸ
ʸ
in its 0,1,2,...,
k
degree neighbors in interval
ʔt
,
LP
would be marked as
TLP
. Algorithm 1,
TPLs
Extraction, illustrates the
procedures of how
TLPs
can be identified.
≥
Algorithm 1.
TLPs
Extraction
Input:
The camera set
CA
, the limit of
k
when searching
k
degree neighbors, the
continuous occurrence limit
ʸ
,andtheinterval
ʔt
between the camera in question
and the
k
degree neighbors
Output:
The
TLPs
of each camera
ca ∈ CA
1:
for
each
ca ∈ CA
do
2:
for
each
LP
recognized by
ca
do
3:
if
ʸ
≥ ʸ
in its 0,1,2,...,
k
degree neighbors in interval
ʔt
then
4:
Mark
LP
as
TLP
;
5:
end if
6:
end for
7:
end for
8:
return
TLPs
Algorithm 1 checks whether
ʸ
can be satisfied first (line 3), if that does at
some
k
i
<k
,
TLPs
Extraction only does calculate camera
ca
sk
i
+1
,k
i
+
2
, ..., k
degree neighbors. Thus it considerably deceases the time complexity of
Algorithm 1. Assume that each camera has
m
1deree neighbors, if we do not
set the constraint on
ʸ
, the time complexity of Algorithm 1 will be
O
nm
k
.
Now under the constraint of
ʸ
, Algorithm 1 breaks the inner loop whenever
ʸ
is
met. Here
ʔt
is used to filter out vehicles that are not the same as the
LP
in
question, based on the assumption that vehicles are floating on the roads.
CTPT Generation (Character Transition Probability Table Genera
tion).
The purpose of the process of CTPT Generation is to generate character
transition probability table which is applied for correcting
SLPs
.Aspointed
out in the beginning of Section 3, we think of the plate as a string. Most ALPR
algorithms recognize each character of the plate independently.
Here, let
Char Set
stand for the set of all characters recognized by ALPR,
i.e.,
Char Set
=
. We can claim that for an incorrect
plate recognized by ALPR, the position of the misrecognized character does not
depend on other characters being correctly recognized or not. For each camera,
a CTPT is built for predicting the error pattern.
Assume that each camera
ca
i
makes mistakes in its own way,
ca
i
∈
{
char
1
,char
2
, ..., char
n
}
CA
,and
there are many external factors affecting recognition result, such as weather
conditions, light intensity, etc. Let
Weather Set
=
stand for va
riety of weather conditions, such as sunny, rainy, foggy, etc. Due to light intensity
changes over time, we divide a day into six time periods for every four hours,
i.e.,
Time Set
=
{
w
1
,w
2
, ..., w
n
}
{
t
1
,t
2
,t
3
,t
4
,t
5
,t
6
}
.Andto
LPs
taken by
ca
i
,
N
(
char
i

w,t
)