Method and apparatus for an adaptive threshold based object detection
View Patent ↗A method, non-transitory computer readable medium, and apparatus for detecting an object in an image are disclosed. For example, the method receives the image, calculates a score for each one of a plurality of locations in the image, performs a box plot of the score of the each one of the plurality of locations of the image, identifies an outlier score that falls outside of the box plot, determines that a distance ratio of the outlier score is less than a predefined distance ratio and detects the object in a location of the plurality of locations of the image corresponding to the outlier score.
1. A method for detecting an object in an image, comprising:
receiving, by a processor, the image;
calculating, by the processor, a score for each one of a plurality of locations in the image;
performing, by the processor, a box plot of the score of the each one of the plurality of locations of the image;
identifying, by the processor, an outlier score that falls outside of the box plot;
determining, by the processor, that a distance ratio of the outlier score is less than a predefined distance ratio, wherein the distance ratio is a ratio that is based upon a comparison of a first distance of a second highest score and a second distance of the outlier score; and
detecting, by the processor, the object in a location of the plurality of locations of the image corresponding to the outlier score.
2. The method of claim 1 , wherein the image is a near infrared (NIR) image of a vehicle.
3. The method of claim 1 , wherein the object is a person and the image is analyzed to determine a total number of people in a vehicle for compliance with a high occupancy vehicle lane or a high occupancy tolling lane.
4. The method of claim 1 , wherein the predefined distance ratio is based upon a plurality of training data.
5. The method of claim 1 , wherein the first distance is measured from a median value of the box plot to the second highest score and the second distance is measured from the median value to the outlier score.
6. The method of claim 1 , wherein the calculating is performed in accordance with a function based upon one or more landmark points and a mixture of parts of the image.
7. The method of claim 6 , wherein the function is defined by an equation comprising:
S
(
I
,
L
,
m
)
=
App
m
(
I
,
L
)
+
Shape
m
(
L
)
+
α
m
,
App
m
(
I
,
L
)
=
∑
i
∈
V
m
w
i
m
·
ϕ
(
I
,
l
i
)
Shape
m
(
L
)
=
∑
ij
∈
E
m
a
ij
m
d
x
2
+
b
ij
m
d
x
+
c
ij
m
d
y
2
+
d
ij
m
d
y
wherein S is a score as a function of the image being analyzed I, the one or more landmark points L tuned for mixture m, φ(I,l i ) is the HoG (Histogram of Gradients) features extracted at location l i , App m is a sum of appearance evidence for placing a template w i m for a part tuned for mixture m at a location l i of the image, Shape m is a score of a mixture specific spatial arrangement of parts L, dx and dy are the x-axis and y-axis displacements of part i with respect to part j, parameters (a, b, c, d) specify spatial cost constraints between pairs of parts i and for a geometric relationship between the one or more landmark points and α m is a constant for the mixture m, wherein V m comprises a pool of parts belonging to the mixture m, wherein E m represents a set of edges between the pool of parts in V m .
8. A non-transitory computer-readable medium storing a plurality of instructions, which when executed by a processor, cause the processor to perform operations for detecting an object in an image, the operations comprising:
receiving the image;
calculating a score for each one of a plurality of locations in the image;
performing a box plot of the score of the each one of the plurality of locations of the image;
identifying an outlier score that falls outside of the box plot;
determining that a distance ratio of the outlier score is less than a predefined distance ratio, wherein the distance ratio is a ratio that is based upon a comparison of a first distance of a second highest score and a second distance of the outlier score; and
detecting the object in a location of the plurality of locations of the image corresponding to the outlier score.
9. The non-transitory computer-readable medium of claim 8 , wherein the image is a near infrared (NIR) image of a vehicle.
10. The non-transitory computer-readable medium of claim 8 , wherein the object is a person and the image is analyzed to determine a total number of people in a vehicle for compliance with a high occupancy vehicle lane or a high occupancy tolling lane.
11. The non-transitory computer-readable medium of claim 8 , wherein the predefined distance ratio is based upon a plurality of training data.
12. The non-transitory computer-readable medium of claim 8 , wherein the first distance is measured from a median value of the box plot to the second highest score and the second distance is measured from the median value to the outlier score.
13. The non-transitory computer-readable medium of claim 8 , wherein the calculating is performed in accordance with a function based upon one or more landmark points and a mixture of parts of the image.
14. The non-transitory computer-readable medium of claim 13 , wherein the function is defined by an equation comprising:
S ( I,L,m )=App m ( I,L )+Shape m ( L )+α m ,
App
m
(
I
,
L
)
=
∑
i
∈
V
m
w
i
m
·
ϕ
(
I
,
l
i
)
Shape
m
(
L
)
=
∑
ij
∈
E
m
a
ij
m
d
x
2
+
b
ij
m
d
x
+
c
ij
m
d
y
2
+
d
ij
m
d
y
wherein S is a score as a function of the image being analyzed I, the one or more landmark points L tuned for mixture m, φ(I,l i ) is the HoG (Histogram of Gradients) features extracted at location l i , App m is a sum of appearance evidence for placing a template w i m for a part tuned for mixture m at a location l i of the image, Shape m is a score of a mixture specific spatial arrangement of parts L, dx and dy are the x-axis and y-axis displacements of part i with respect to part j, parameters (a, b, c, d) specify spatial cost constraints between pairs of parts i and j or a geometric relationship between the one or more landmark points and α m is a constant for the mixture m, wherein V m comprises a pool of parts belonging to the mixture m, wherein E m represents a set of edges between the pool of parts in V m .
15. A method for detecting an object in an image, comprising:
analyzing, by a processor, a plurality of training images having a known object to calculate a predefined distance ratio;
receiving, by the processor, the image, wherein the image is of a vehicle;
calculating, by the processor, a score for each one of a plurality of locations in the image, wherein the scoring is performed in accordance with a function based upon one or more landmark points and a mixture of parts of the image;
performing, by the processor, a box plot of the score of the each one of the plurality of locations of the image, the box plot comprising a lowest score, a 25 th percentile, a median, a 75 th percentile and a second highest score;
identifying, by the processor, an outlier score that falls outside of the box plot;
determining, by the processor, that a distance ratio of the outlier score is less than the predefined distance ratio;
detecting, by the processor, the object in a location of the plurality of locations of the image corresponding to the outlier score, wherein the object is a person; and
calculating, by the processor, a total number of objects in the image to determine if the vehicle is in violation of a total number of passengers requirement for a high occupancy vehicle lane.
16. The method of claim 15 , wherein the image is a near infrared (NIR) image.
17. The method of claim 15 , wherein the image is captured when the vehicle enters the high occupancy vehicle lane.
18. The method of claim 15 , wherein the function is defined by an equation comprising:
S
(
I
,
L
,
m
)
=
App
m
(
I
,
L
)
+
Shape
m
(
L
)
+
α
m
,
App
m
(
I
,
L
)
=
∑
i
∈
V
m
w
i
m
·
ϕ
(
I
,
l
i
)
Shape
m
(
L
)
=
∑
ij
∈
E
m
a
ij
m
d
x
2
+
b
ij
m
d
x
+
c
ij
m
d
y
2
+
d
ij
m
d
y
wherein S is a score as a function of the image being analyzed I, the one or more landmark points L tuned for mixture m, φ(I,l i ) is the HoG (Histogram of Gradients) features extracted at location l i , App m is a sum of appearance evidence for placing a template w i m for a part tuned for mixture m at a location l i of the image, Shape m is a score of a mixture specific spatial arrangement of parts L, dx and dy are the x-axis and y-axis displacements of part i with respect to part j, parameters (a, b, c, d) specify spatial cost constraints between pairs of parts i and j or a geometric relationship between the one or more landmark points and α m is a constant for the mixture m, wherein V m comprises a pool of parts belonging to the mixture m, wherein E m represents a set of edges between the pool of parts in V m .