text
stringlengths 0
89.3k
|
|---|
occlusion is different for each frame centralization for the
|
observation matrix Wdoes not guarantee that the transla
|
tion between shapes is eliminated Therefore recentering
|
of the shapes is required before alignment using the TPA
|
module which only requires modification of S
|
SiRpiSiT i 1 F 37
|
where the definition of TI1
|
P11Tis translation re
|
moval matrix The completed model after adding occlusion
|
is as followsmin
|
SQµ1
|
2OWΠS2
|
Fµ2
|
S
|
µ3
|
2F1X
|
i1
|
QiSiQi1Si1
|
2
|
F
|
st
|
SgˆSΛ
|
ˆSiQiSi i 1 F
|
SiRpiSiT i 1 F38
|
The update formulas for variables Ωcan be obtained by im
|
itating the above solution procedure Since the improve
|
ment of the model is only related to the optimization vari
|
ableSS we only need to adjust their update formulas
|
Solution for Sunder occlusion Sin model 38 needs to
|
be centralized and then transformed by rotation Rpito get
|
S Thus we only need to replace Sin Eq 31 with STie
|
µ3
|
βQTHTHQ 2I3FS
|
QTˆS1
|
βQTY2RpST1
|
βY339
|
Solution for Sunder occlusion We solve the 3D shape Si
|
framebyframe and the optimization model is as follows
|
Si arg min
|
Siµ1
|
2OiWiΠiSi2
|
F
|
β
|
2
|
SiRpiSiT
|
2
|
FD
|
Yi
|
3SiRpiSiTE
|
40
|
where Oi12oi We define MidiagoiRPP
|
then model 40 can be rewritten as
|
Si arg min
|
Siµ1
|
2WiΠiSiMi2
|
F
|
β
|
2
|
SiRpiSiT
|
2
|
FD
|
Yi
|
3SiRpiSiTE
|
41
|
Calculating the derivative of model 41 and equating it to
|
zero yields the solution of Sias follows
|
µ1
|
βΠT
|
iΠiSiM2
|
iSiT2
|
µ1
|
βΠT
|
iWiM2
|
iRT
|
piSiT1
|
βRT
|
piYi
|
3T42
|
In order to calculate the closedform solution of Si we use
|
the property vecAXB BTAvecXto equiva
|
lently represent Eq 42 as
|
µ1
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.