bundle adjustment and slam - cvg @ ethz

3D Photography: Bundle Adjustment and SLAM

31 March 2014

1

Structure-From-Motion

• Two views initialization:

– 5-Point algorithm (Minimal Solver)

– 8-Point linear algorithm

– 7-Point algorithm

E → (R,t)

2


• Triangulation: 3D Points

E → (R,t)

3


• Subsequent views: Perspective pose estimation

(R,t) (R,t)

(R,t) 4

Bundle Adjustment

• Final step in Structure-from-Motion.

• Refine a visual reconstruction to produce jointly optimal 3D structures P and camera poses C.

• Minimize total reprojection errors .

ijx

i j

WijijX ij

CPx2

,argmin

ijz

z

jP

jj CX ,

iC

ijz

Cost Function:

Measurement error covariance :1

ijW

],[ CPX 5

Bundle Adjustment

• Final step in Structure-from-Motion.

• Refine a visual reconstruction to produce jointly optimal 3D structures P and camera poses C.

• Minimize total reprojection errors .

ijx

z

jP

jj CX ,

iC

ijz

Cost Function:

i j

ijij

T

ijX

zWzargmin

Measurement error covariance :1

ijW

],[ CPX 6

Xf

Bundle Adjustment

• Minimize the cost function:

1. Gradient Descent

2. Newton Method

3. Gauss-Newton

4. Levenberg-Marquardt

XfX

argmin

7

Bundle Adjustment

1. Gradient Descent

8

Initialization: 0XX k

Compute gradient:

gXX kk

WJZ

X

Xfg T

XX k

Update:

Slow convergence near minimum point!

Iterate until convergence

: Step size X

J

: Jacobian

Bundle Adjustment

2. Newton Method

9

2nd order approximation (Quadratic Taylor Expansion):

KK XX

T

XXHgXfXf

21)()(

Find that minimizes ! KXX

Xf

)(

kXX

XfH

2

2

:

Hessian matrix

Bundle Adjustment

2. Newton Method

10

Differentiate and set to 0 gives:

gH 1

Computation of H is not trivial and might get stuck at saddle point!

Update: kk XX

Bundle Adjustment

3. Gauss-Newton

11

i j

ij

ijij

T

XWZWJJH

2

2

WJJH T

Normal equation:

ZWJWJJ TT

Update: kk XX

Might get stuck and slow convergence at saddle point !

Bundle Adjustment

12

4. Levenberg-Marquardt

Regularized Gauss-Newton with damping factor .

ZWJIWJJ TT

:0 Gauss-Newton (when convergence is rapid)

: Gradient descent (when convergence is slow)

LMH

Structure of the Jacobian and Hessian Matrices

• Sparse matrices since 3D structures are locally observed.

13

Solving the Normal Equation

• Schur Complement

14

ZWJH T

LM

LMH

3D Structures

Camera Parameters



15

ZWJH T

LM

3D Structures

Camera Parameters

sH

LMH

SCH

T

SCH CH



16

ZWJH T

LM

C

S

C

S

C

T

SC

SCS

HH

HH

3D Structures

Camera Parameters

Multiply both sides by:

IHH

I

S

T

SC

1

0

110 S

T

SCSC

S

C

S

SCS

T

SCC

SCS

HHHHHH

HH



17

110 S

T

SCSC

S

C

S

SCS

T

SCC

SCS

HHHHHH

HH

11 )( S

T

SCSCCSCS

T

SCC HHHHHH

First solve for from: C

Schur Complement (Sparse and Symmetric Positive Definite Matrix)

Easy to invert a block diagonal matrix

Solve for by backward substitution. SC


• Sparse matrix factorization 1. LU Factorization

2. QR factorization

3. Cholesky Factorization

• Iterative methods 1. Conjugate gradient

2. Gauss-Seidel

18

11 )( S

T

SCSCCSCS

T

SCC HHHHHH bAx

Can be solved without inverting A since it is a sparse matrix!

QRA

LUA

TLLA

Solve for x by forward backward substitutions.

Problem of Fill-In

19

Problem of Fill-In

• Reorder sparse matrix to minimize fill-in.

• NP-Complete problem.

• Approximate solutions: 1. Minimum degree

2. Column approximate minimum degree permutation

3. Reverse Cuthill-Mckee.

4. …

20

bPxPAPP TTT

Permutation matrix to reorder A

Problem of Fill-In

21

Robust Cost Function

• Non-linear least squares:

• Maximum log-likelihood solution:

• Assume that: 1. X is a random variable that follows Gaussian distribution.

2. All observations are independent.

22

ij

ijij

T

ijX

zWzargmin

)|(lnargmin- XZpX

ij

ijij

T

ijijXX

zWzcZXp explnargmin-)|(lnargmin-

ij

ijij

T

ijX

zWzargmin


• Gaussian distribution assumption is not true in the presence of outliers!

• Causes wrong convergences.

23


24

ij

ijij

T

ijXij

ijijX

zSzz argminargmin

Robust Cost Function scaled with ijW ij"

• Similar to iteratively re-weighted least-squares.

• Weight is iteratively rescaled with the attenuating factor .

• Attenuating factor is computed based on current error.

ij"


25

(.) (.)"

Reduced influence from high errors

Gaussian Distribution

Cauchy Distribution

Influence from high errors


26

Outliers are taken into account in Cauchy!

State-of-the-Art Solvers

• Google Ceres:

– https://code.google.com/p/ceres-solver/

• g2o:

– https://openslam.org/g2o.html

• GTSAM:

– https://collab.cc.gatech.edu/borg/gtsam/

27

https://code.google.com/p/ceres-solver/



https://openslam.org/g2o.html

https://collab.cc.gatech.edu/borg/gtsam/

Simultaneous Localization and Mapping (SLAM)

• For a robot to estimate its own pose and acquire a map model of its environment.

• Chicken-and-Egg problem:

– Map is needed for localization.

– Pose is needed for mapping.

28

Full SLAM: Problem Definition

29

Robot Poses

Observations

Map landmarks

M

i

K

k

jkikkiiiX,LX,L

lxzpuxxpXpUZLXp1 1

10 ),|(),|()(argmax,|,argmax

u1 u2 u3 Control Actions

30

M

i

K

k

jkikkiiiX,LX,L

lxzpuxxpXpUZLXp1 1

10 ),|(),|()(argmax,|,argmax

K

k

jkikk

M

i

iiiX,L

lxzpuxxp11

1 ),|(ln),|(lnargmin-Negative log-likelihood


Likelihoods:

}),(exp{),|(2

11i

iiiii xuxfuxxp

Process model

}),(exp{),|(2

kkjkikjkikk zlxhlxzp

Measurement model


31

K

k

jkikk

M

i

iiiX,LX,L

lxzpuxxpUZLXp11

1 ),|(ln),|(lnargmin-),|,(argmax

Putting the likelihoods into the equation:

K

k

kjkik

M

i

iiiX,LX,L ki

zlxhxuxfUZLXp1

2

1

2

1 ),(),(argmin),|,(argmax

Minimization can be done with Levenberg-Marquardt (similar to bundle adjustment)!


32

ZWJIWJJ TT

Jacobian made up of , , , x

f

u

f

x

h

l

h

Weight made up of , i

k

Normal Equations:

Can be solved with sparse matrix factorization or iterative methods

Online SLAM: Problem Definition

33

121 ...),|,(),|,( tt dxdxdxUZLXpUZLxp

• Estimate current pose and full map .

• Inference with:

1. Kalman Filtering (EKF SLAM)

2. Particle Filtering (FastSLAM)

tx L

Previous poses are marginalized out

EKF SLAM

• Assumes: pose and map are random variables that follows Gaussian distribution.

• Hence,

34

tx L

),Ν(~),|,( UZLxp t

mean Error covariance

EKF SLAM

35

),( 1 ttt uf

t

T

tttt RFF 1

1)( t

T

ttt

T

ttt QHHHK

tttt yK

tttt HKI )(

1

1),(

t

ttt

x

ufF

t

tt

x

hH

)(

Prediction:

Correction:

Process model

Error propagation with process noise

Kalman gain

Update mean

)( ttt hzy Measurement residual (innovation)

Update covariance

Measurement Jacobian Process Jacobian

Structure of Mean and Covariance

36

2

2

2

2

2

2

2

1

21

2221222

1211111

21

21

21

,

NNNNNN

N

N

N

N

N

llllllylxl

llllllylxl

llllllylxl

lllyx

ylylylyyxy

xlxlxlxxyx

t

N

t

l

l

l

y

x

Covariance is a dense matrix that grows with increasing map features!

True robot and map states might not follow unimodal Gaussian distribution!

Particle Filtering: FastSLAM

• Particles represents samples from the posterior distribution .

• can be any distribution (need not be Gaussian).

37

),|,( UZLxp t

),|,( UZLxp t

FastSLAM

38

},...,,,,{ ,,,2,2,1,1 m

tN

m

tN

m

t

m

t

m

t

m

t

m

t

m

t xp

Each particle represents:

Robot state Landmark state (mean and covariance)

),|(~ 1 ttt

m

t uxxpx Sample the robot state from

the process model

),|( , t

m

t

m

tn zxLp N Kalman filter measurement updates

),|( m

t

m

tt

m

t xLzpw Weight update

Resampling

FastSLAM

• Many particles needed for accurate results.

• Computationally expensive for high state dimensions.

39

Pose-Graph SLAM

40

ij

jiijX ij

vvhz2

),(argmin

• 3D structures are removed (fixed).

• Constraints are relative pose estimates from 3D structures.

• Minimizes loop-closure errors.

Conclusion

• Bundle Adjustment

• Simultaneous Localization and Mapping

– Full SLAM

– Online SLAM

• EKF SLAM

• FastSLAM

• Pose-Graph SLAM

41

bundle adjustment and slam - cvg @ ethz

Documents