Summary of important papers for the lecture "Non-Rigid-Deformation"

forpointclouds
bae correspondence optimisationfornonrigidregistration ofDepthscans slap
Glo

forpartialscansofdeformingshapesthataddressesthechallengesofnonrigidregistrationwithinasinglenon
algorithm
linear
optimization
registration

This
algorithm wasfor
simultaneouslyso thereliabilityofeach
correspondencesconfidenceweightsthat
measure c orrespondence

and
identifiesnonoverlappingareasandawarping that
field thesourcescanintoalignmentwith
brings thetargetgeometry
local
poor minimaare avoidedwithaniterativeexecutionscheme thatdetectssuboptimaeconvergenceand the
restarts
theoptimizationwithimprovedinitialconditionssothatbetterresultisobtained
employsanon
Thealgorithm systemthat
lineardeformation favorsnaturaldeformationsbymaximizingboth
rigidity
andconsistency
Theoptimizationoswessimultaneouslyforboththedeformationparametersaswellasthecorrespondentpositions
sincesome sourcepointshave nocorrespondingpositiononthetargetduetopartialoverlapweaugment
eachcorrespondencewithaweightthatisalsosowedforbythe optimizer


Deformationmodel


sourcescanisaugmentedwitharedureddeformablemodelintheformof a graph deformation

Esmoon serves as aregularizesforthedeformationbyindicatingthatthe
affinetransformations
ofadjacentgraphnodesshouldagreewithoneanother
erigid penalizesthedeviationofeachtransformationfromapurerigidmotion
consequentlylocalfeatures
deformasrigidly aspossible shearingorstretchingartifacts
avoiding
strivestomoveeachsourcegraphnodetoits correspondingpositiononthe
targetshape
exit
Esont aimsatmaximizingthenumber
maximizestheregion
ofreliablecorrespondencesbypushingtheweightstowardsonethus
ofoverlap



RealtimenonrigidreconstructionusinganRab Dcamera cmapsoobjecttopointcloud

I create a JD model
2 nonrigid registrationofliveratsDdatatothe smoothtemplate
usinganextendednonlinearasrigidas possibleCARAP framework

realtimenonrigidreconstruction

rigidregistration roughlyalignsthetemplatetotheinputdata
2 Nonrigidsurfacefitting
minimization
offittingenergywhichcombinesdensegeometricandphotometricmodeltodata
constraints APregularizer
aswellasanasrigidaspossiblecar
Gauss usingthe
solver
Newton preconditionedconjugate method
gradient spca inits
Theenergyisminimizedusinga
innerloop
hierarchypreparedattemplate
Thissolverisappliedinacoarse
acquisition
tofinemannerusingthemultiresolution
mesh
ateachlevelthe energyis
fitting optimized atthe
currentresolution iterations GaussNewton
several using of
pointsu to so
image mapsas
depth apoint
won
ri acairn aruleirs
epointinus apointyni o od nisvisibility
associated with
each everse
mode

theenergy od of
estimate sensor
noise
inamain theclosestairfare
point searchwithin gives srobustwind
acomplicatedenergy so lift the innerminimise X matrix
sacooian
tobecomesearchparameterso suis towards
etommnoimqjaownn.eea

eplaneinus inane zip ncant cri
on daily
thisimprovingconvergence
t ermallows a
incorrectly ssigned toslidealongthesurface
correspondences
thus
fromsocoimagecoordinate
asprojection
orinis acolor nip Ii Is corny
it
eco Ii It i cacintin colorattachedtoeach from
vertex pretime step
as
rearsaniaatto nwin signsee t.in
emaireg the
sowing line

sis
so
not
me compare
to nwe at
seaward
use him
no asi n
entry
iniq and
eaionoen
update

erne y linearoranaa forsmallenames
ratio soon
parameter are inan
anat estimated initial
roster
forlargere
linearorconstant