19
Multiple Frame Motion Inference Using Belief Propagation Jiang Gao Jianbo Shi Presented By: Gilad Kapelushnik Visual Recognition, Spring 2005, Technion IIT.

Multiple Frame Motion Inference Using Belief Propagation

Embed Size (px)

DESCRIPTION

Multiple Frame Motion Inference Using Belief Propagation. Jiang Gao Jianbo Shi. Presented By: Gilad Kapelushnik. Visual Recognition, Spring 2005, Technion IIT. Abstract. S4(X,Y). Find “best fit” upper body joint configuration. Input is a 2D video - PowerPoint PPT Presentation

Citation preview

Page 1: Multiple Frame Motion Inference Using Belief Propagation

Multiple Frame Motion Inference Using Belief

Propagation

Jiang Gao Jianbo Shi

Presented By: Gilad Kapelushnik

Visual Recognition, Spring 2005, Technion IIT.

Page 2: Multiple Frame Motion Inference Using Belief Propagation

Abstract

• Find “best fit” upper body joint configuration.

• Input is a 2D video

• Each joint is described by its location on a 2D grid.

S1(X,Y)

S2(X,Y)

S4(X,Y)

S6(X,Y)

S5(X,Y)

S3(X,Y)

Let J be a joint configuration – {S1,S2,S3,S4,S5,S6}

We would like to find:

argmax ( | )J

J P J observations

Page 3: Multiple Frame Motion Inference Using Belief Propagation

• Step 1: Subtract two sequential frames.

• Step 2: Apply threshold.

Motion Energy Image

Page 4: Multiple Frame Motion Inference Using Belief Propagation

From #NrgPixels To Probability• Sum the Energy Pixels in the Patch.

• Calculate probability using the following:

S5(10,60)

S6(40,30)

1 exp( # )

NormConstP

NrgPixels

Page 5: Multiple Frame Motion Inference Using Belief Propagation

0.12

0.19

0.84

0.68

0.02

• Find configuration J with the highest probability.

• Computing all possible probabilities is inefficient.

• a-Priori data give better and faster results.

• removing impossible configurations reduce inference time.

Main Idea

( )JX Y

Page 6: Multiple Frame Motion Inference Using Belief Propagation

a-Priori Data• A probability table for Each P(Sx,Sy).

• Compute probability at grid crossing.

• Use nearest neighbor for the rest of the image.

Example:

For right arm - P(S2,S3)

Red – Low probability

Green – High probability

P(S2,S3)12…Ns^2

100…0

200.1…0

…………0

Ns^20000

Page 7: Multiple Frame Motion Inference Using Belief Propagation

• Face is detected using face detection algorithm.

• Initial assumption of Shoulders from face and pose.

• Even using BP there are too many possible states to go through.

• Candidates for elbows from shoulders & Energy Map.

• Candidates for Wrists from skin color model.

Detect Candidate states (1)

Page 8: Multiple Frame Motion Inference Using Belief Propagation

Detect Candidate states (2)• Many states can be discarded.

• Remove close candidate states.

Pros: Much faster inference.

Cons: Less accurate.

• Note: This is only an option.

Fits skin color and wrist location

Pink for right wristRed for left wrist

Blue for elbow

Page 9: Multiple Frame Motion Inference Using Belief Propagation

The Markov Model• Empty Circles - States - 2D positions of joints

• Full Circles - Observations - Computed from energy map.

• Each state correspond to an observation.

Page 10: Multiple Frame Motion Inference Using Belief Propagation

Belief Propagation (1)• Solve inference problem using an algorithm with Linear complexity.

• Each joint has a vector with probabilities for each candidate.

Shoulder

Elbow

Wrist

Page 11: Multiple Frame Motion Inference Using Belief Propagation

Belief Propagation (2)

1

2

3

m23

m32

m21

m12m14

m41For each iteration:

• Each node sends a message to its neighbor nodes containing the “wanted” probability (for each state).

• Messages are computed according to:

( ) /

( )

( , ) ( | ) ( )

( ) ( | ) ( )

i

ij ij i j i i i ki is k N i j

i i i i i ki ik N i

m P s s P x s m s

b s P x s m s

Sum over all candidates

A-priori Data for each state.

Normalize variable.

Observation (# of Energy pixels in patch) for each state

converted to a probability.

Message from k to i (all messages from the

neighbors). This is actually a vector with a probability for

each state.

Message from i to j.

Page 12: Multiple Frame Motion Inference Using Belief Propagation

Belief Propagation (3) - Example

21

Message from 1 to 2

4states

2states

Page 13: Multiple Frame Motion Inference Using Belief Propagation

Belief Propagation (4)

• BP converge after 2-4 iterations (giving the right a-Priori data).

• For every joint there is a probability vector for each candidate state.

Page 14: Multiple Frame Motion Inference Using Belief Propagation

Multiple Frame Probability

• Multiple frame (8) is proposed for smoother transition between configurations.

• Prevents joints changing their state to a different which is “far away” (Euclidian distance).

• Though BP was designed to work with loopy-free models, the author stated that it worked fine.

21

, 1 12

1( , ) exp

2

t ti it t t t

i i i

s sP s s

And for those who really want to know:

Page 15: Multiple Frame Motion Inference Using Belief Propagation

2D to 3D

• 2D -> 3D by Taylor (2000).

• Assuming (u1,v1) and (u2,v2) are projections then depth can be retrieved using the following:

1 2 1 2

1 2 1 2

2 22 1 2 1 2

1 2 2

( ) ( )

( ) ( )

( ) ( ){ _ }

co

u u const X X

v v const Y Y

u u v vZ Z arm length

nst

Page 16: Multiple Frame Motion Inference Using Belief Propagation

Results(1)

Page 17: Multiple Frame Motion Inference Using Belief Propagation

Results(2)

Page 18: Multiple Frame Motion Inference Using Belief Propagation

Results(3)

Errors accrue when 2 joints intersect each other.

On some occasions, even when limbs intersect, it was possible to infer correctly.

Page 19: Multiple Frame Motion Inference Using Belief Propagation

Q?