Linear system based, qualitative independent motion detection from compressed MPEG surveillance video
View Patent ↗The present invention features a qualitative method to detect independent motion revealed in successive frames of a compressed surveillance MPEG video stream using linear system consistency analysis without decompression of the stream, identifying the segments containing independent motion in a real-time or faster manner, for the retrieval of these segments. The linear system is constructed using the macroblocks of MPEG compressed video frames. The normal flow value of the macroblock is obtained by taking the dot product between the macroblock gradient vector, computed by averaging the four block gradient vectors, and the motion vector of this macroblock. The normal flow value is filtered for inclusion in the linear system, and the statistic of the matrices of the resulting linear system is determined, filtered to screen out false negatives and outliers, and used to determine the presence or absence of independent motion.
1. A method of qualitative detection of independent motion in a pair of video frames, the steps comprising:
a) building with an automated data processor a linear system representing a three dimensional scene, for a pair of two dimensional video frames, wherein a three-dimensional to two-dimensional affine transform is employed in which parallel lines remain parallel after transformation, and a change in camera position with respect to objects represented in the pair of video frames does not render the linear system unsolvable;
b) automatically analyzing with an automated data processor said linear system to determine the solvability thereof, substantially without quantitatively solving said linear system; and
c) when said analyzing step (b) determines said linear system to be unsolvable, indicating that one of said pair of video frames contains independent motion relative to the other of said pair of video frames,
wherein the solvability of the linear system is determined based on a respective rank of a matrix D and an augmented matrix Db, wherein:
D
ξ
=
b
D
=
(
1
y
.
1
x
1
y
1
⋯
⋯
⋯
⋯
1
y
.
n
x
n
y
n
)
ξ
=
(
θ
α
β
γ
)
b
=
(
-
x
1
⋯
-
x
n
)
wherein α, β, γ and θ are functions of three-dimensional motion parameters between the pair of frames and the three-dimensional to two-dimensional affine transform parameters which represent the parameters of a linear constraint
{dot over (x)}+θ{dot over (y)}+αx+βy+γ= 0
for each image point of a respective video frame p at coordinates (x, y) having a time derivative ({dot over (x)}, {dot over (y)}).
2. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 1 , wherein said indicating step (c) comprises labeling at least one of said pair of video frames as containing independent motion.
3. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 1 , wherein said analyzing step (b) comprises performing a consistency analysis of said linear system.
4. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 1 , wherein said pair of video frames is selected from a stream of compressed video data and said building a linear system step (a), said analyzing step (b) and said indicating step (c) are performed in at least real time upon said pairs of frames of said compressed video data.
5. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 4 , wherein said stream of compressed video data comprises MPEG-encoded video data.
6. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 5 , wherein said pair of video frames are MPEG encoded and comprise pixels, blocks comprising arrays of said pixels, and macroblocks comprising arrays of said blocks.
7. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 6 , wherein said analyzing step (b) comprises estimating a motion vector from at least one of said macroblocks.
8. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 7 , wherein said analyzing step (b) comprises the sub-step: i) estimating a motion vector from at least one of said macroblocks.
9. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 8 , wherein said estimating sub-step (i) comprises estimating said motion vector using a normal flow computation.
10. The method according to claim 1 , wherein the solvability of the linear system is determined substantially without spatial segmentation in the image domain.
11. The method according to claim 1 ,
wherein said automatically analyzing comprises comparing a rank of matrix D with a rank of augmented matrix Db, by applying Singular Value Decomposition to both matrix D and augmented matrix Db, defining
R
=
σ
min
(
D
)
σ
min
(
Db
)
,
where σ min (D) and σ min (Db) stand for the smallest singular values of D and Db, respectively, and wherein a solvability of the linear system is concluded if R is above a predetermined threshold.
12. A real-time method of qualitative detection of independent motion in a stream of compressed video data, the steps comprising:
a) selecting a pair of two-dimensional video frames from a stream of compressed video data representing a three dimensional scene;
b) building a linear system representing the three-dimensional scene for said pair of two-dimensional video frames with an automated data processor, wherein a three-dimensional to two-dimensional affine transform is employed in which parallel lines remain parallel after transformation, and a change in camera position with respect to objects represented in the pair of video frames does not render the linear system unsolvable;
c) automatically analyzing said linear system with an automated data processor to determine the solvability thereof substantially without quantitatively solving said linear system; and
d) when said analyzing step (c) determines said linear system to be unsolvable, indicating that one of said pair of video frames contains independent motion relative to the other of said pair of video frames,
wherein the solvability of the linear system is determined based on a respective rank of a matrix D and an augmented matrix Db, wherein:
D
ξ
=
b
D
=
(
1
y
.
1
x
1
y
1
⋯
⋯
⋯
⋯
1
y
.
n
x
n
y
n
)
ξ
=
(
θ
α
β
γ
)
b
=
(
-
x
1
⋯
-
x
n
)
wherein α, β, γ and θ are functions of three-dimensional motion parameters between the pair of frames and the three-dimensional to two-dimensional affine transform parameters which represent the parameters of a linear constraint
{dot over (x)}+θ{dot over (y)}+αx+βy+γ= 0
for each image point of a respective video frame p at coordinates (x, y) having a time derivative ({dot over (x)}, {dot over (y)}).
13. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 12 , wherein said analyzing step (c) comprises performing a consistency analysis of said linear system.
14. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 12 , wherein said stream of compressed video data comprises MPEG-encoded video data.
15. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 14 , wherein said MPEG-encoded video data comprises pixels, blocks comprising arrays of said pixels, and macroblocks comprising arrays of said blocks.
16. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 15 , wherein said analyzing step (c) comprises estimating a motion vector from at least one of said macroblocks.
17. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 16 , wherein said analyzing step (c) comprises the sub-step: i) estimating a motion vector from at least one of said macroblocks.
18. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 17 , wherein said estimating sub-step (i) comprises estimating said motion vector using a normal flow computation.
19. The method according to claim 12 ,
wherein said automatically analyzing comprises comparing a rank of matrix D with a rank of augmented matrix Db, by applying Singular Value Decomposition to both matrix D and augmented matrix Db, defining
R
=
σ
min
(
D
)
σ
min
(
Db
)
,
where σ min (D) and σ min (Db) stand for the smallest singular values of D and Db, respectively, and wherein a solvability of the linear system is concluded if R is above a predetermined threshold.
20. The method of qualitative detection of independent motion in a pair of video frames as recited in claim 12 , wherein the solvability of the linear system is determined substantially without spatial segmentation in the image domain.
21. An apparatus for qualitative detection of independent motion in a pair of images, comprising:
a) an input receiving data representing a pair of two dimensional images;
b) an automated data processor, adapted to:
i) build a linear system representing a three dimensional scene, based on the received two dimensional image data, wherein a three dimensional to two dimensional affine transform is employed in which parallel lines remain parallel after transformation, and a change in camera position with respect to objects represented in the three dimensional scene does not render the linear system unsolvable, and
ii) determine a solvability of the linear system substantially without quantitatively solving the linear system; and
c) an output adapted to communicate an indication when said computer determines the built linear system to be unsolvable, that one of said pair of video frames contains independent motion relative to the other of said pair of video frames
wherein n points are identified in a first video frame, and each point's respective displacement in the second video frame is ({dot over (x)}, {dot over (y)}) T , each point contributing one linear constraint of
{dot over (x)}+θ{dot over (y)}+αx+βy+γ= 0,
wherein α, β, γ and θ are functions of three-dimensional motion parameters between the pair of frames and the three-dimensional to two-dimensional affine transform parameters for each image point p at coordinates (x, y) having a time derivative ({dot over (x)}, {dot over (y)}), wherein the automated data processor determines a solvability of the linear system:
Dξ=b
where
D
=
(
1
y
.
1
x
1
y
1
⋯
⋯
⋯
⋯
1
y
.
n
x
n
y
n
)
ξ
=
(
θ
α
β
γ
)
b
=
(
-
x
1
⋯
-
x
n
)
.
22. The apparatus according to claim 21 , wherein the automated data processor, determines a solvability of the linear system based on a statistic of the linear system substantially without quantitatively solving the linear system.
23. The apparatus according to claim 21 , wherein the automated data processor determines solvability of the linear system substantially without spatial segmentation in the image domain.
24. The apparatus according to claim 21 , wherein the automated data processor determines the solvability by comparing a rank of matrix D with a rank of augmented matrix Db, by applying Singular Value Decomposition to both matrix D and augmented matrix Db, defining
R
=
σ
min
(
D
)
σ
min
(
Db
)
,
where σ min (D) and σ min (Db) stand for the smallest singular values of D and Db, respectively, and wherein a consistency of the linear system is concluded if R is above a predetermined threshold.
25. The apparatus according to claim 21 , wherein said pair of images are represented within MPEG-encoded video data.
26. The apparatus according to claim 25 , wherein said MPEG-encoded video data comprises pixels, blocks comprising arrays of said pixels, and macroblocks comprising arrays of said blocks, and wherein the automated data processor estimates a motion vector from at least one of said macroblocks.