bmc 2012
TRANSCRIPT
Foreground Detection via Robust Low Rank MatrixDecomposition including spatio-temporal Constraint
C. Guyon, T. Bouwmans and E. Zahzah
MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France—
Workshop BMC, Daejong Korea
November 5, 2012
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 1 / 27
Summary
1 Introduction and motivations on IRLS
2 Temporal constraint with an adapted Norm
3 Diagram flow and Spatial constraint
4 Experimental Results
5 Conclusion
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 2 / 27
Introduction and motivations
PurposeForeground detection : Segmentation of moving objects in videosequence acquired by a fixed camera.Background modeling : Modelization of all that is not movingobject.
Involved applicationsSurveillance cameraMotion capture
On the importanceCrucial Task : Often the first step of a full video surveillance system.
Strategy usedEigenbackground decomposition.
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 3 / 27
Eigenbackgrounds
Find an « ideal » subspace of the video sequence, which describes the bestas possible the (dynamic) background.
Fig.1 The common process of background subtraction via PCA (Principal ComponentAnalysis). At final step, an adaptative threshold is used to get a binary image.
Without a robust framework, the moving object may be absorbed in the model !C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 4 / 27
Data Structure Transformation
First, we consider a video sequence as a matrix A ∈ Rn×m
n is the amount of pixels in a frame (∼ 106)m is the number of frames considered (∼ 200)
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 5 / 27
IRLS : Vector version (1)
The usual IRLS (Iteratively reweighted least squares) scheme for solveargmin
x||Ax − b||α is given by
D(i) = diag((ε+ |b − Ax (i)|)α−2)x (i+1) = (AtD(i)A)−1AtD(i)b
(1)
that a suitable IRLS method is convergent for 1 ≤ α < 3Other formulation,
r (i) = b − Ax (i)
D = diag((ε+ |r (i)|)α−2)y (i) = (A′DA)−1A′Dr (i)
x (i+1) = x (i) + (1+ λopt)y (i)
(2)
λopt computed in a second inner loop and convergent for 1 ≤ α < +∞
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 6 / 27
IRLS : Vector version (2)
For Spatio/temporal RPCA, needs to solve the following general problem :
argminx
||Ax − b||α + λ||Cx − d ||β (3)
By derivation, the associated IRLS scheme is,
r1 = b − Ax (i), r2 = d − Cx (i), e1 = ε+ |r1|, e2 = ε+ |r2|D1 = (
∑eα1 )
1α−1diag(eα−21 ),D2 = λ(
∑eβ2 )
1β−1diag(eβ−22 )
y (i) = (A′D1A + C ′D2C)−1(A′D1r1 + C ′D2r2)
x (i+1) = x (i) + (1 + λopt)y (i)
(4)
Good news : Just few line of matlab code !
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 7 / 27
Matrix Version
More generally, we consider the following matrix regression problem withtwo parameters norm (α, β) and a weighted matrix (W ),
minX||AX − B||α,β
Wwith ||Mij ||α,β
W= (
n∑i=1
(m∑
j=1
Wij |Mij |β)αβ )
1α (5)
The problem is solved in the same manner on matrices with a reweightedregression strategy,
Until X is stable, repeat on each k-columnsR ← B − AXS ← ε+ |R|Dk ← diag(Sβ−2ik ◦ (
∑j (S
βij ◦Wij ))
αβ−1 ◦Wik)k
Xik ← Xik +(1+Λ(max(α, β)))(AtDkA)−1AtDkRik
(6)
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 8 / 27
Exemple : Geometric median (convex problem)
Fig.1 Find the x-axis points which minimize the ||.||α distance between the 7 otherspoints (black stars). Minimum of Curves is corresponding to this optimal value for α =2, 1.66, 1.33 and 1 (blue, green, red, black).
Fig.1 Same in Two Dimension. By continuously changing the minization problem (i.e.varying α) during each iterations, this trick accelerates the convergence of IRLS.
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 9 / 27
Various RPCA formulation (only for α = 1)
PCA with fixed rank is : minL,S
||S ||Fs.t. Rank(L) = k
A = L + S(7)
R(obust)PCA is (Non convex and NP-hard) :
minL,S
||σ(L)||0 + λ||S ||0s.t. A = L + S
(8)
Convex relaxed problem of (2) is RPCA-PCP proposed by Candès et al. [1] :
minL,S
||σ(L)||1 + λ||S ||1s.t. A = L + S
(9)
Where σ(L) means singular values of L.A mix is Stable PCP of Zhou et al. [2] (both entry-wise and sparse noise) :
minL,S
||σ(L)||1 + λ||S ||1s.t. ||A− L− S ||F < δ
(10)
All of them could be solved by Augmented Lagrangian Multipliers (ALM).
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 10 / 27
Video examples
Some examples, temporal RPCA and ideal RPCA with groundtruth fittingand optimal parameters fitting.
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 11 / 27
Summary
1 Introduction and motivations on IRLS
2 Temporal constraint with an adapted Norm
3 Diagram flow and Spatial constraint
4 Experimental Results
5 Conclusion
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 12 / 27
Sparse solution
In RPCA, residual error is sparse.Using the RPCA decomposition on a synthetic low-rank random matrixplus noise, the error looks like :
Same principle with video. Sparse noise (or outliers) are the moving objects.
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 13 / 27
Let’s play with norms
Varying the α, β norm → Different kind of recovering pattern error.
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 14 / 27
Let’s play with norms...(2)
Some issuesWhat is the best specific norm for temporal constrain ?Initial assumption is ||.||2,1. Confirmed experimentally ?
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 15 / 27
Validation
If ideal eigenbakgrounds are that, best norm must be ...
Let’s denote Lopt , the ideal low-rank subspace which outliers do not contribute to PCA
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 16 / 27
Experimental results
Let’s denote Lα,β , the low-rankrecovered matrix with a ||.||α,β-PCA.The plot show the error between||Lopt − Lα,β||F for parameters αand β chosen freely. The darkestvalue means that the error is thesmallest here.
||S ||2,1 is not optimal, but for convenience we use it.The benefit of the ad hoc block-sparse hypothesis is confirmed bytesting its efficiency directly on video dataset.
Experimentation done on dynamic category of dataset change detectionworkshop 2012 : http://www.changedetection.net/
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 17 / 27
Summary
1 Introduction and motivations on IRLS
2 Temporal constraint with an adapted Norm
3 Diagram flow and Spatial constraint
4 Experimental Results
5 Conclusion
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 18 / 27
Overview and addition of the spatial constraint
Figure: Overview of the learning and evaluation process. Learning process needsGT (Groundtruth) for better fits the eigenbackground components.
Spatial ConstrainSuppose A = L+ S and L and S are computed via some kind ofRPCA technique with the addtion of Total Variation penalty on S .This increase connected (or connexe) shapes.
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 19 / 27
Exemple with a synthetic 1-D signal
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 20 / 27
Summary
1 Introduction and motivations on IRLS
2 Temporal constraint with an adapted Norm
3 Diagram flow and Spatial constraint
4 Experimental Results
5 Conclusion
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 21 / 27
Experimental Protocol
Optimal threshold is chosen for maximizing F-measure criterionwhich is based 2× 2 histogram of True/false/positive/negative :
DR =TP
TP + FN, Prec =
TP
TP + FP, F =
2 DR PrecDR + Prec
Good performance is then obtained when the F-measure is closed to 1Time consumption is not take into account in the evaluation process.
RPCA-LBD is compared with the following two Robust methods :Robust Subspace Learning (RSL) De La Torre et al. [4]Principal Component Pursuit (RPCA-PCP) [1]
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 22 / 27
Quantitative Results
Here, we show experimental results on the real dataset of BMC,
Video Recall Precision F-measure PSNR Visual Results1 0.9139 0.7170 0.8036 38.24252 0.8785 0.8656 0.8720 26.77213 0.9658 0.8120 0.8822 37.70534 0.9550 0.7187 0.8202 39.36995 0.9102 0.5589 0.6925 30.58766 0.9002 0.7727 0.8316 29.99947 0.9116 0.8401 0.8744 26.83508 0.8651 0.6710 0.7558 30.50409 0.9309 0.8239 0.8741 55.1163
Table: Quantitative results with common criterions. Last column show theoriginal, GT and result of the first four real video sequences.
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 23 / 27
Other results
Figure: First eigenBackground of the fifths sequence of Rotary (BMC) with thenorms ||.||opt , ||.||1,1 and ||.||2,1. Last row shows the first five eigenBackground onreal dataset with ||.||2,1
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 24 / 27
Summary
1 Introduction and motivations on IRLS
2 Temporal constraint with an adapted Norm
3 Diagram flow and Spatial constraint
4 Experimental Results
5 Conclusion
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 25 / 27
Qualitative analysis
AdvantagesExperiments on video surveillance datasets show that this approach ismore robust than RSL and RPCA-PCP in presence of dynamicbackgrounds and illumination changes.Well suited for video with spatially spread and temporarily sparseoutliers.
DisadvantagesNot efficient on sequences with outliers always presents.For small local variations, like « wind tree » are not (yet) wellmodelized by this kind global PCA. Probably needs more eigencomponents !
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 26 / 27
Future Works & References
Future WorksLack in computation time : Further research consists in developping anincremental version to update the model at every frame and to achievethe real-time requirements.
References[1] E. Candes, X. Li, Y. Ma, and J. Wright, Robust principal component
analysis, International Journal of ACM, vol. 58, no. 3, May 2011.[2] Z. Zhou, X. Li, J. Wright, E. Candes, and Y. Ma, Stable principal
component pursuit,IEEE ISIT Proceedings, pp. 1518-1522, Jun. 2010.[3] G. Tang and A. Nehorai, Robust principal component analysis based
on low-rank and block-sparse matrix decomposition, CISS 2011, 2011.
C. Guyon, T. Bouwmans and E. Zahzah (MIA Laboratory (Mathematics Images & Applications), University of La Rochelle, France — Workshop BMC, Daejong Korea )Foreground Detection via Robust Low Rank Matrix Decomposition including spatio-temporal ConstraintNovember 5, 2012 27 / 27