Some - - PowerPoint PPT Presentation

▶
some of the figures and slides are adapted from s sinha j
SMART_READER_LITE
LIVE PREVIEW

Some - - PowerPoint PPT Presentation

Nov 19, 2022 34 likes •579 views

Some of the figures and slides are adapted from S. Sinha, J.S. Franco, J. Matusiks presentations, and referenced papers.

slide-1
SLIDE 1
  • Some of the figures and slides are adapted from S. Sinha, J.S. Franco, J. Matusik’s

presentations, and referenced papers.

slide-2
SLIDE 2
slide-3
SLIDE 3
  • Silhouettes are the regions where objects of

interest project in images

  • Silhouettes can generally be obtained using
  • Silhouettes can generally be obtained using

low level information (fast)

  • They give information about the global shape
  • f scene objects
slide-4
SLIDE 4
  • Sometimes done manually (for offline

applications, ground truth and verifications)

  • Region based-extraction (automatic)
  • silhouette extraction is a 2-region image
  • silhouette extraction is a 2-region image

segmentation problem, w/ specific solutions:

  • chroma keying (blue, green background)
  • background subtraction (pre-observed static or

dynamic background)

slide-5
SLIDE 5
  • Contour-based extraction
  • focus on silhouette outline instead of region

itself

  • snakes, active contours: fitting of a curve to high

gradients in image, local optimization

Yilmaz&Shah ACCV04

slide-6
SLIDE 6
  • Background subtraction
  • !
  • "#$%&'#
  • (
  • (
  • #))
  • *'

+!,

  • ,

, +!

slide-7
SLIDE 7
slide-8
SLIDE 8

'.,

slide-9
SLIDE 9

)'.,/ (

  • Maximal volume consistent

with silhouettes

[Laurentini94] [Baumgart74]

  • Can be seen as the

Visual hull

  • Can be seen as the

intersection of viewing cones

Viewing cone

  • Properties:
  • Containment property: contains real scene objects
  • Converges towards the shape of scene objects minus

concavities as N increases

  • Projective structure: simple management of visibility

problems

slide-10
SLIDE 10

0,(

# 1, *',2

  • ,,3
slide-11
SLIDE 11

(/-%&4

slide-12
SLIDE 12
slide-13
SLIDE 13

Voxel grid Surface A priori knowledge Voxel grid

Volumetric approaches

Surface

Surface approaches

A priori knowledge ex: articulated model Polyhedron mesh

Image-based approaches

slide-14
SLIDE 14

#

5(/

../ 6.

5&

./, .

5&

#/

., .

!

slide-15
SLIDE 15
  • .

4, 7'.

slide-16
SLIDE 16

(.

7,%& ..

..

? /, /+ ? ?

slide-17
SLIDE 17

1../

  • .

[Szeliski TR 90’]

slide-18
SLIDE 18

(

8%&. 4.. ,..

slide-19
SLIDE 19
  • .

4, 7'.

slide-20
SLIDE 20

49,)

8 .,. '. 4!%, , 5& , ,.

  • W. Matusik, An Efficient Visual Hull Computation Algorithm
slide-21
SLIDE 21

(

  • /

! !, , !

,.

slide-22
SLIDE 22

(

:'';

2, ,. 2

Boyer IJCV 95

slide-23
SLIDE 23

(

/,

Boyer IJCV 95

slide-24
SLIDE 24

(9,

,

  • $<=>?@/
  • .
slide-25
SLIDE 25

(

  • AB!C</,.,

9/,

slide-26
SLIDE 26

(

  • AB!C</,.,

9/, Rim mesh Rims Visual hull Strips

slide-27
SLIDE 27

49,(

D8E$, $)(1 5CC%F

  • ,

. ,

Triple point Viewing edges Cone intersection edges Discrete visual hull

,

  • Cone intersection edges

Viewing cone

  • The contour discretization imposes the geometry of viewing

cones, and the geometry of the surface

  • viewing edges
  • cone intersection edges
slide-28
SLIDE 28

49(-

3 steps:

  • 1. Compute viewing edges
  • 2. Cone intersection edges and triple points
  • 3. Faces.
slide-29
SLIDE 29

1,

7,. ,',

  • . /

GB. + 975&%&

slide-30
SLIDE 30

7.

,. ,. 4.,!' %&, 4,

  • .,,

. ! ,,

[W. Matusik, Image Based Visual Hulls]

slide-31
SLIDE 31

(.'

  • /,!
  • /.'

Matusik&al Siggraph 00

slide-32
SLIDE 32

7'$1

Reference 1 Reference 2 Desired

slide-33
SLIDE 33

.

7

slide-34
SLIDE 34
  • .',
slide-35
SLIDE 35

(,-

slide-36
SLIDE 36

(,5&

Desired view Reference view

slide-37
SLIDE 37

(,5&

Desired view Reference view

Front-most Points

slide-38
SLIDE 38

(,5&

Desired view Reference view Visible

slide-39
SLIDE 39

(,5&

Desired view Reference view

Coverage Mask

slide-40
SLIDE 40

(,5&

Desired view Reference view

Coverage Mask

Visible

slide-41
SLIDE 41

(,5&

Desired view Reference view

Coverage Mask

Visible

slide-42
SLIDE 42

(,5&

Desired view Reference view

Coverage Mask

Visible Not

slide-43
SLIDE 43

(,5&

Desired view Reference view

Coverage Mask

slide-44
SLIDE 44

(,5&

Desired view Reference view

Coverage Mask

Visible

slide-45
SLIDE 45

(,5&

Desired view Reference view

Coverage Mask

slide-46
SLIDE 46

(,5&

Desired view Reference view

Coverage Mask

Visible Not

slide-47
SLIDE 47

7$("

  • ,
  • 9B

1 1

slide-48
SLIDE 48

7$("

Movie

slide-49
SLIDE 49
slide-50
SLIDE 50

0, ,

slide-51
SLIDE 51

0, ,

  • Idea: we wish to find the content of the scene from images, as a

probability grid

??

probability grid

  • Modeling the forward problem - explaining image observations

given the grid state - is easy. It can be accounted for in a sensor model.

  • Bayesian inference enables the formulation of our initial inverse

problem from the sensor model

  • Simplification for tractability: independent analysis and

processing of voxels

slide-52
SLIDE 52

7

  • Unreliable silhouettes: do not make decision about their

location

  • Do sensor fusion: use all image information simultaneously
slide-53
SLIDE 53

)

Sensor model: Grid

  • I: color information in images
  • B: background color model
  • F: silhouette detection variable (0 or 1): hidden
  • OX: occupancy at voxel X (0 ou 1)

∑

=

F X X

F P B F I P O I P ) O | ( ) , | ( ) | (

Inference:

∑ ∏ ∏

=

X

O pixel img X pixel img pixel img X pixel img X

I P I P I O P

, , , ,

) O | ( ) O | ( ) | (

slide-54
SLIDE 54

(B