Graphical object models for detection and tracking
View Patent ↗A computer implemented method for object detection includes providing a spatio-temporal model for an object to be detected, providing a video including a plurality of images including the object, and measuring the object as a collection of components in each image. The method further includes determining a probability that the object is in each image, and detecting the object in any image upon comparing the probabilities for each image to a threshold for detecting the object.
1. A computer implemented method for object detection comprising:
providing a spatio-temporal model for an object to be detected;
providing a video comprising a plurality of images including the object;
measuring the object as a collection of components in each image;
determining a probability that the object is in each image; and
detecting the object in any image upon comparing the probabilities for each image to a threshold for detecting the object.
2. The computer implemented method of claim 1 , wherein providing the spatio-temporal model for the object to be detected comprises providing detectors for each of the collection of components.
3. The computer implemented method of claim 1 , wherein the spatio-temporal model is a graphical model comprising nodes corresponding to each of the collection of components and to the object.
4. The computer implemented method of claim 1 , wherein determining the probability that the object is in each image comprises detecting the object in a current image according to measurements of the object as a collection of components determined from a prior image and a later image relative to the current image.
5. The computer implemented method of claim 1 , wherein providing the spatio-temporal model for the object to be detected further comprises providing a temporal window defining a plurality of images in which measurements of components detected therein are passed to components detected in the current image.
6. The computer implemented method of claim 1 , wherein determining the probability that the object is in each image comprises determining the probability for a position and a size of the object in each image.
7. The computer implemented method of claim 1 , wherein the threshold is provided for the object to be detected, wherein the threshold is determined empirically.
8. The computer implemented method of claim 1 , wherein a joint probability distribution for the spatio-temporal model with N components is:
P
(
X
0
O
,
X
0
C
0
,
X
0
C
1
,
…
,
X
0
C
N
,
…
,
X
T
O
,
X
T
C
0
,
X
T
C
1
,
…
,
X
T
C
N
)
=
1
Z
∏
ij
ψ
ij
(
X
i
O
,
X
j
O
)
∏
ik
ψ
ik
(
X
i
O
,
X
i
C
k
)
∏
ikl
ψ
kl
(
X
i
C
k
,
X
i
C
l
)
∏
i
ϕ
i
(
X
i
O
)
∏
ik
ϕ
i
(
X
i
C
k
)
.
9. A computer readable medium embodying a computer program to perform method steps for object detection, the method steps comprising:
providing a spatio-temporal model for an object to be detected;
providing a video comprising a plurality of images including the object;
measuring the object as a collection of components in each image;
determining a probability that the object is in each image; and
detecting the object in any image upon comparing the probabilities for each image to a threshold for detecting the object.
10. The method of claim 9 , wherein providing the spatio-temporal model for the object to be detected comprises providing detectors for each of the collection of components.
11. The method of claim 9 , wherein the spatio-temporal model is a graphical model comprising nodes corresponding to each of the collection of components and to the object.
12. The method of claim 9 , wherein determining the probability that the object is in each image comprises detecting the object in a current image according to measurements of the object as a collection of components determined from a prior image and a later image relative to the current image.
13. The method of claim 9 , wherein providing the spatio-temporal model for the object to be detected further comprises providing a temporal window defining a plurality of images in which measurements of components detected therein are passed to components detected in the current image.
14. The method of claim 9 , wherein determining the probability that the object is in each image comprises determining the probability for a position and a size of the object in each image.
15. The method of claim 9 , wherein the threshold is provided for the object to be detected, wherein the threshold is determined empirically.
16. The method of claim 9 , wherein a joint probability distribution for the spatio-temporal model with N components is:
P
(
X
0
O
,
X
0
C
0
,
X
0
C
1
,
…
,
X
0
C
N
,
…
,
X
T
O
,
X
T
C
0
,
X
T
C
1
,
…
,
X
T
C
N
)
=
1
Z
∏
ij
ψ
ij
(
X
i
O
,
X
j
O
)
∏
ik
ψ
ik
(
X
i
O
,
X
i
C
k
)
∏
ikl
ψ
kl
(
X
i
C
k
,
X
i
C
l
)
∏
i
ϕ
i
(
X
i
O
)
∏
ik
ϕ
i
(
X
i
C
k
)
.