30
Galilean Differential Geometry of Moving Images Daniel Fagerström CVAP/NADA/KTH [email protected]

Galilean Differential Geometry Of Moving Images

Embed Size (px)

Citation preview

Page 1: Galilean Differential Geometry Of Moving Images

Galilean Differential Geometry of Moving Images

Daniel FagerströmCVAP/NADA/[email protected]

Page 2: Galilean Differential Geometry Of Moving Images

Differential Structure of Movies

• How can we describe the local structure of an image sequence?

• We will assume that a movie is a smooth function of 2+1 dimensional space-time

• Looking for generic properties

Page 3: Galilean Differential Geometry Of Moving Images

Approaches for Motion Analysis

• Optical flow• Spatio-temporal texture• Spatio-temporal differential invariants

Page 4: Galilean Differential Geometry Of Moving Images

Optical Flow

• Geometry of the projected motion of particles in the observers field of view

• Binding hypothesis needed to use it on image sequences

• “Top down”• Local formulation, but non-local, due to binding

hypothesis• Undefined when particles appears and disappears,

e.g. motion boundaries• Does not use image structure

Page 5: Galilean Differential Geometry Of Moving Images

Spatio-Temporal Differential Invariants

• Local geometry of spatio-temporal images• ”Bottom Up”, low level• No binding hypotheses, connection to the

environment considered as a higher level problem

• Well defined everywhere• Does use image structure, extension of low

level vision for still images

Page 6: Galilean Differential Geometry Of Moving Images

Overview

• Galilean geometry• Moving frames• Image geometry• 1+1 dimensional Galilean differential invariants• 2+1 dimensional Galilean differential invariants• What is required for a more realistic movie model

Page 7: Galilean Differential Geometry Of Moving Images

Galilean Geometry

• Spatial and temporal translation a, spatial rotation R and spatio-temporal shear v

nSORvaatxvR

tx n

,,,10

x

t

x

t

Shear

Page 8: Galilean Differential Geometry Of Moving Images

Galilean Geometry

• Insensitive to constant relative motion for parallel projection, approximately otherwise

• Simplest meaning full model• Assumed implicitly when one talk about optical

flow invariants: div, rot, dev, i.e. first order flow • Shape properties from the environment can be

derived from relative motion• (Newton physics describe Galilean invariants)

Page 9: Galilean Differential Geometry Of Moving Images

Galilean Invariants

• Planes of simultaneity (constant t) are invariant and has Euclidean geometry: distances and angles are invariants– i.e. an image sequence

• The temporal distance between planes of simultaneity is an invariant

0 if,,

,,,

tyxp

tyxptp

S

T

Page 10: Galilean Differential Geometry Of Moving Images

Galilean ON-System

• An n+1 dimensional Galilean ON-system (e1,e2,,e0) is s.t. (e1,e2,,en) is an Euclidean ON-system and ||e0||T=1

Page 11: Galilean Differential Geometry Of Moving Images

Moving Frames

• Galilean geometry has no metric• We will use Cartan's method of moving

frames, that does not require a metric• Moving frame: e:M! G½ GL(n)• Attach a frame that is adapted to the local

structure in each point• Differential geometry: the local change of

the frame: de

Page 12: Galilean Differential Geometry Of Moving Images

Moving Frames

1

1

ABACACABC

eACedAAdAide

Aie

C(A) contains the differential geometric invariants expressed in the global frame i

Page 13: Galilean Differential Geometry Of Moving Images

Image Geometry

• Image space: E2­I - trivial fiber bundle with Euclidian base space and log intensity as fiber (Koenderink 02) z=f(x,y)

• f is smooth• Image geometry

– Global gray level transformations– Lightness gradients

Page 14: Galilean Differential Geometry Of Moving Images

Gradient Gauge

• For points where rf0 we can choose an adapted ON-frame {u,v} s.t. fu=0

• All functions over iuj

v f, i+j¸ 1 becomes invariants w.r.t. rotation in space and translation in intensity

Aiffff

f y

x

yx

xy

v

u

1

Page 15: Galilean Differential Geometry Of Moving Images

Gray Level Invariants

tionstransformaintensity monotonicand spacein rotation w.r.t.invariant

curvaturegradient -curvature isophote -

,

,0

0

12

12

12

2212

12

12

dvdudvffdu

ffc

cc

ffdyffffdxffff

c

cc

AC

v

uv

v

uu

v

u

yx

xyyyyxxxyxyx

0,0

0,0

Page 16: Galilean Differential Geometry Of Moving Images

Hessian Gauge

• For points whererf0 we can choose an ON-frame {p,q} s.t. fpq=0 and |fpp|>|fqq|

• All functions over ipj

q f, i+j¸ 2 becomes invariants w.r.t. rotation in space, translation in intensity and addition of a linear light gradient (Koenderink 02)

xxyy

xy

y

x

q

p

fff

Ai

2tan,cossinsincos

Page 17: Galilean Differential Geometry Of Moving Images

Galilean 1+1 D

• Two cases:– Isophotes cut the spatial line, motion according

to the constant brightness assumption– Isophotes are tangent to the spatial line (along

curves), creation, annihilation

Page 18: Galilean Differential Geometry Of Moving Images

Tangent Gauge

• Let {t,x} be a global Galilean ON-frame, for points where fx0 we can define an adapted Galilean ON-frame {s,x} s.t. fs=0.

.10

1

,0,

Aiff

fff

x

txt

x

s

xts

xts

Page 19: Galilean Differential Geometry Of Moving Images

Isophote Invariants

divergence- onaccelerati-

,00

0

x01

s01

01

01

cca

dxdsa

dxffds

ffc

cAC

x

sx

x

ss

0,0

a

0,0

a

Page 20: Galilean Differential Geometry Of Moving Images

Hessian Gauge

• Let {t,x} be a global Galilean ON-frame, for points where fxx0 we can define an adapted Galilean ON-frame {r,x} s.t. frx=0.

.10

1

,0,

Aiff

ffff

x

txxtx

x

r

xxtxxrrx

xtr

Page 21: Galilean Differential Geometry Of Moving Images

Hessian Invariants

divergence- onaccelerati-

,00

0

x01

r01

01

01

cca

dxdra

dxffds

ffc

cAC

xx

rxx

xx

rrx

0,0

a

0,0

a

Page 22: Galilean Differential Geometry Of Moving Images

Galilean 2+1 D

• General case• Also here are two different main cases

– Isophote surfaces transversal to the spatial plane. Motion of isophote curves in the image

– Isophote surfaces tangent to the plane. Creation, annihilation and saddle points

Page 23: Galilean Differential Geometry Of Moving Images

Invariants in the General Case

au, av - accelerationu, v - divergenceu, v - skew of the ”flow field” - rotation of the plane in the temporal directionu, v - flow line curvature in the plane

∂s

∂u

∂v=1 v x v y

0 cos −sin 0 sin cos ∂t

∂x

∂ y=Ai ,

C A=0 au dsu du u dv av ds v duv dv0 0 ds u duv dv0 −ds u du v dv 0

Page 24: Galilean Differential Geometry Of Moving Images

More Descriptive Invariants

• D - rate of strain tensor for the spatio-temporal part of the frame field

• a, curl D, div D, def D - are flow field invariants• a, , , u, v - are not flow field invariants

a=au2av2 , a=arctan av /au ,

D=u u

v v =u−v

2 0 1−1 0uv

2 1 00 11

2 u−v uv

u v v−u = curlD

2 0 1−1 0divD

2 1 00 1defD

2Q −11 0

0 −1Q

Page 25: Galilean Differential Geometry Of Moving Images

Tangent Gauge

• Let {t, x, y} be a global Galilean ON-frame, for points where ||{fx,fy}||0 we can define an adapted Galilean ON-frame {s,u,v} s.t. fs=fu=fsu=0

• Principal acceleration extrema

• Direction of u constant along s – used in Guichard (98)

Page 26: Galilean Differential Geometry Of Moving Images

Tangent Gauge

1

22

100010

1

00

00

0011

BABCBCBAC

BAiff

ff

ffff

fffffff

Aiffff

ff

v

u

tv

t

uu

tu

uuv

uvt

v

u

s

uvuutusu

vts

vuts

y

x

t

yx

xy

yxv

u

t

uu

suv

uuv

uvsv

vsvv

uusuuu

vssv

uu

ssu

uuv

uvssu

vv

uu

ff

ffff

ffff

ffaff

ffffa

dvducdvdsac

dvdudsac

cccc

BAC

12

02

01

12

12

0201

0000

0

Page 27: Galilean Differential Geometry Of Moving Images

Hessian Gauge

• Let {t, x, y} be a global Galilean ON-frame, we define an adapted Galilean ON-frame {r, p, q} s.t. fpq= frp= frq=0.

• Also defined when the spatial tangent disappears, e.g. for creation and disappearance of structure

• r is the same vector field as when the optical flow constraint equation is solved for the spatial image gradient

Page 28: Galilean Differential Geometry Of Moving Images

Hessian Gauge

1

22

2

2tan

cossin0sincos0001

1

100010

11

0

0

BABCBCBAC

fff

BAiff

Aiffffffff

fff

ffff

ffff

xxyy

xy

y

x

r

yxq

p

r

y

x

txxtyxytxyytxxyty

xyyyxxy

x

r

yyxytyry

xyxxtxrx

yxtr

)22/(

0000

0

12

02

01

12

12

0201

qqpprpq

qqrpqq

pprpqp

qqrqqq

pprppp

qqrrqq

pprrpp

qp

qqq

ppp

fff

ff

ff

ff

ff

ffa

ffa

dqdpc

dqdpdrac

dqdpdrac

cccc

BAC

Page 29: Galilean Differential Geometry Of Moving Images

Real Image Sequences

• Localized filters are not invariant w.r.t. Galilean shear, velocity adapted (s – directed) filters are needed

• More generic singular cases for imprecise measurements

Page 30: Galilean Differential Geometry Of Moving Images

Conclusion

• Theory about differential invariants for smooth Galilean spatio-temporal image sequences

• Local operators• “Bottom up”• Contains more information about the image

sequence than optical flow• Extension of methods for still images