- ORB SLAM3系统初始化
- ORB SLAM3 构建Frame
- ORB_SLAM3 单目初始化
- ORB_SLAM3 双目匹配
- ORB_SLAM3_IMU预积分理论推导(预积分项)
- ORB_SLAM3_IMU预积分理论推导(噪声分析)
- ORB_SLAM3_IMU预积分理论推导(更新)
- ORB_SLAM3_IMU预积分理论推导(残差)
- ORB_SLAM3_优化方法 Pose优化
- ORB_SLAM3 闭环检测
噪声分析
-
δ
ϕ
⃗
i
j
\delta \vec{\phi}_{i j}
δϕij:
Exp ( − δ ϕ ⃗ i j ) = ∏ k = i j − 1 Exp ( − Δ R ~ k + 1 j T ⋅ J r k ⋅ η k g d Δ t ) \operatorname{Exp}\left(-\delta \vec{\phi}_{i j}\right)=\prod_{k=i}^{j-1} \operatorname{Exp}\left(-\Delta \tilde{\mathbf{R}}_{k+1 j}^{T} \cdot \mathbf{J}_{r}^{k} \cdot \boldsymbol{\eta}_{k}^{g d} \Delta t\right) Exp(−δϕij)=k=i∏j−1Exp(−ΔR~k+1jT⋅Jrk⋅ηkgdΔt)
其中, J r k = J r ( ( ω ~ k − b i g ) Δ t ) \mathbf{J}_{r}^{k}=\mathbf{J}_{r}\left(\left(\tilde{\boldsymbol{\omega}}_{k}-\mathbf{b}_{i}^{g}\right) \Delta t\right) Jrk=Jr((ω~k−big)Δt)
对上式两边取对数:
δ
ϕ
⃗
i
j
=
−
log
(
∏
k
=
i
j
−
1
Exp
(
−
Δ
R
~
k
+
1
j
T
⋅
J
r
k
⋅
η
k
g
d
Δ
t
)
)
\delta \vec{\phi}_{i j}=-\log \left(\prod_{k=i}^{j-1} \operatorname{Exp}\left(-\Delta \tilde{\mathbf{R}}_{k+1 j}^{T} \cdot \mathbf{J}_{r}^{k} \cdot \boldsymbol{\eta}_{k}^{g d} \Delta t\right)\right)
δϕij=−log(k=i∏j−1Exp(−ΔR~k+1jT⋅Jrk⋅ηkgdΔt))
利用性质:当
δ
ϕ
⃗
\delta \vec{\phi}
δϕ是小量时,
log
(
Exp
(
ϕ
⃗
)
⋅
Exp
(
δ
ϕ
⃗
)
)
=
ϕ
⃗
+
J
r
−
1
(
ϕ
⃗
)
⋅
δ
ϕ
⃗
\log{\left(\operatorname{Exp}\left(\vec{\phi} \right) \cdot \operatorname{Exp}\left(\delta\vec{\phi}\right)\right)}=\vec{\phi }+\operatorname{J}^{-1}_{r}\left (\vec{\phi } \right ) \cdot \delta \vec{\phi }
log(Exp(ϕ)⋅Exp(δϕ))=ϕ+Jr−1(ϕ)⋅δϕ, 其中
η
k
g
d
\boldsymbol{\eta}_{k}^{g d}
ηkgd是小量,
ξ
=
Δ
R
~
k
+
1
j
T
⋅
J
r
k
⋅
η
k
g
d
Δ
t
\xi =\Delta \tilde{\mathbf{R}}_{k+1 j}^{T} \cdot \mathbf{J}_{r}^{k} \cdot \boldsymbol{\eta}_{k}^{g d} \Delta t
ξ=ΔR~k+1jT⋅Jrk⋅ηkgdΔt是小量,于是
J
r
−
1
(
ξ
k
)
≈
I
\operatorname{J}^{-1}_{r}\left (\xi_{k} \right )\approx I
Jr−1(ξk)≈I
δ
ϕ
⃗
i
j
=
−
log
(
∏
k
=
i
j
−
1
Exp
(
−
ξ
k
)
)
=
−
log
(
Exp
(
−
ξ
i
)
∏
k
=
i
+
1
j
−
1
Exp
(
−
ξ
k
)
)
≈
−
(
−
ξ
i
+
I
⋅
log
(
∏
k
=
i
+
1
j
−
1
Exp
(
−
ξ
k
)
)
)
=
ξ
i
−
log
(
∏
k
=
i
+
1
j
−
1
Exp
(
−
ξ
k
)
)
=
ξ
i
−
log
(
Exp
(
−
ξ
i
+
1
)
∏
k
=
i
+
2
j
−
1
Exp
(
−
ξ
k
)
)
≈
ξ
i
+
ξ
i
+
1
−
log
(
∏
k
=
i
+
2
j
−
1
Exp
(
−
ξ
k
)
)
≈
⋯
≈
∑
k
=
i
j
−
1
ξ
k
\begin{aligned} \delta \vec{\phi}_{i j} & =-\log \left(\prod_{k=i}^{j-1} \operatorname{Exp}\left(-\xi_{k}\right)\right) \\ & =-\log \left(\operatorname{Exp}\left(-\xi_{i}\right) \prod_{k=i+1}^{j-1} \operatorname{Exp}\left(-\xi_{k}\right)\right) \\ & \approx-\left(-\xi_{i}+\mathbf{I} \cdot \log \left(\prod_{k=i+1}^{j-1} \operatorname{Exp}\left(-\xi_{k}\right)\right)\right)=\xi_{i}-\log \left(\prod_{k=i+1}^{j-1} \operatorname{Exp}\left(-\xi_{k}\right)\right) \\ & =\xi_{i}-\log \left(\operatorname{Exp}\left(-\xi_{i+1}\right) \prod_{k=i+2}^{j-1} \operatorname{Exp}\left(-\xi_{k}\right)\right) \\ & \approx \xi_{i}+\xi_{i+1}-\log \left(\prod_{k=i+2}^{j-1} \operatorname{Exp}\left(-\xi_{k}\right)\right) \\ & \approx \cdots \\ & \approx \sum_{k=i}^{j-1} \xi_{k} \end{aligned}
δϕij=−log(k=i∏j−1Exp(−ξk))=−log(Exp(−ξi)k=i+1∏j−1Exp(−ξk))≈−(−ξi+I⋅log(k=i+1∏j−1Exp(−ξk)))=ξi−log(k=i+1∏j−1Exp(−ξk))=ξi−log(Exp(−ξi+1)k=i+2∏j−1Exp(−ξk))≈ξi+ξi+1−log(k=i+2∏j−1Exp(−ξk))≈⋯≈k=i∑j−1ξk
即:
δ
ϕ
⃗
i
j
≈
∑
k
=
i
j
−
1
Δ
R
~
k
+
1
j
T
J
r
k
η
k
g
d
Δ
t
\delta\vec{\phi}_{ij}\approx \sum_{k=i}^{j-1} {\Delta\tilde{R}_{k+1 j}^{T}\operatorname{J}_{r}^{k}\eta_{k}^{gd}\Delta t }
δϕij≈k=i∑j−1ΔR~k+1jTJrkηkgdΔt
由于
Δ
R
~
k
+
1
j
T
\Delta\tilde{R}_{k+1 j}^{T}
ΔR~k+1jT、
J
r
k
\operatorname{J}_{r}^{k}
Jrk、
Δ
t
\Delta t
Δt都是已知量,而
η
k
g
d
\eta_{k}^{gd}
ηkgd为零均值高斯噪声,因此
δ
ϕ
⃗
i
j
\delta \vec{\phi}_{i j}
δϕij也为零均值高斯噪声
2.
δ
v
i
j
\delta v_{ij}
δvij
δ
v
i
j
=
∑
k
=
i
j
−
1
[
Δ
R
~
i
k
η
k
a
d
Δ
t
−
Δ
R
~
i
k
⋅
(
f
~
k
−
b
i
a
)
∧
⋅
δ
ϕ
⃗
i
k
⋅
Δ
t
]
\delta \mathbf{v}_{i j}=\sum_{k=i}^{j-1}\left[\Delta \tilde{\mathbf{R}}_{i k} \mathbf{\eta}_{k}^{a d} \Delta t-\Delta \tilde{\mathbf{R}}_{i k} \cdot\left(\tilde{\mathbf{f}}_{k}-\mathbf{b}_{i}^{a}\right)^{\wedge} \cdot \delta \vec{\phi}_{i k} \cdot \Delta t\right]
δvij=k=i∑j−1[ΔR~ikηkadΔt−ΔR~ik⋅(f~k−bia)∧⋅δϕik⋅Δt]
3.
δ
p
i
j
\delta p_{ij}
δpij
δ
p
i
j
=
∑
k
=
i
j
−
1
[
δ
v
i
k
Δ
t
−
1
2
Δ
R
~
i
k
⋅
(
f
~
k
−
b
i
a
)
∧
δ
ϕ
⃗
i
k
Δ
t
2
+
1
2
Δ
R
~
i
k
η
k
a
d
Δ
t
2
]
\delta \mathbf{p}_{i j}=\sum_{k=i}^{j-1}\left[\delta \mathbf{v}_{i k} \Delta t-\frac{1}{2} \Delta \tilde{\mathbf{R}}_{i k} \cdot\left(\tilde{\mathbf{f}}_{k}-\mathbf{b}_{i}^{a}\right)^{\wedge} \delta \vec{\phi}_{i k} \Delta t^{2}+\frac{1}{2} \Delta \tilde{\mathbf{R}}_{i k} \mathbf{\eta}_{k}^{a d} \Delta t^{2}\right]
δpij=k=i∑j−1[δvikΔt−21ΔR~ik⋅(f~k−bia)∧δϕikΔt2+21ΔR~ikηkadΔt2]
噪声更新
-
δ
ϕ
→
i
j
−
1
→
δ
ϕ
→
i
j
\delta\overrightarrow{\phi}_{ij-1}\to \delta\overrightarrow{\phi}_{ij}
δϕij−1→δϕij
δ ϕ ⃗ i j = ∑ k = i j − 1 Δ R ~ k + 1 j T J r k η k g d Δ t = ∑ k = i j − 2 Δ R ~ k + 1 j T J r k η k g d Δ t + Δ R ~ j j T ⏟ I J r j − 1 η j − 1 g d Δ t = ∑ j − 2 k = i ( Δ R ~ k + 1 j − 1 Δ R ~ j − 1 j ) T J r k η k g d Δ t + J r j − 1 η j − 1 g d Δ t = Δ R ~ j j − 1 ∑ k = i j − 2 Δ R ~ k + 1 j − 1 T J r k η k g d Δ t + J r j − 1 η j − 1 g d Δ t = Δ R ~ j j − 1 δ ϕ ⃗ i j − 1 + J r j − 1 η j − 1 g d Δ t \begin{aligned} \delta \vec{\phi}_{i j} & =\sum_{k=i}^{j-1} \Delta \tilde{\mathbf{R}}_{k+1 j}^{T} \mathbf{J}_{r}^{k} \boldsymbol{\eta}_{k}^{g d} \Delta t \\ & =\sum_{k=i}^{j-2} {\color{Red} \Delta \tilde{\mathbf{R}}_{k+1 j}^{T}} \mathbf{J}_{r}^{k} \boldsymbol{\eta}_{k}^{g d} \Delta t+\underbrace{\Delta \tilde{\mathbf{R}}_{j j}^{T}}_{I} \mathbf{J}_{r}^{j-1} \boldsymbol{\eta}_{j-1}^{g d} \Delta t \\ &= \sum_{j-2}^{k=i}{\color{Red} \left(\Delta \tilde{\mathbf{R}}_{k+1 j-1} \Delta \tilde{\mathbf{R}}_{j-1 j}\right)^{T}} \mathbf{J}_{r}^{k} \boldsymbol{\eta}_{k}^{g d} \Delta t+\mathbf{J}_{r}^{j-1} \boldsymbol{\eta}_{j-1}^{g d} \Delta t \\ & =\Delta \tilde{\mathbf{R}}_{j j-1} \sum_{k=i}^{j-2} \Delta \tilde{\mathbf{R}}_{k+1 j-1}^{T} \mathbf{J}_{r}^{k} \boldsymbol{\eta}_{k}^{g d} \Delta t+\mathbf{J}_{r}^{j-1} \boldsymbol{\eta}_{j-1}^{g d} \Delta t \\ & =\Delta \tilde{\mathbf{R}}_{j j-1} \delta \vec{\phi}_{i j-1}+\mathbf{J}_{r}^{j-1} \boldsymbol{\eta}_{j-1}^{g d} \Delta t \end{aligned} δϕij=k=i∑j−1ΔR~k+1jTJrkηkgdΔt=k=i∑j−2ΔR~k+1jTJrkηkgdΔt+I ΔR~jjTJrj−1ηj−1gdΔt=j−2∑k=i(ΔR~k+1j−1ΔR~j−1j)TJrkηkgdΔt+Jrj−1ηj−1gdΔt=ΔR~jj−1k=i∑j−2ΔR~k+1j−1TJrkηkgdΔt+Jrj−1ηj−1gdΔt=ΔR~jj−1δϕij−1+Jrj−1ηj−1gdΔt -
δ
v
i
j
−
1
→
δ
v
i
j
\delta v_{ij-1} \to \delta v_{ij}
δvij−1→δvij
δ v i j = ∑ k = i j − 1 [ Δ R ~ i k η k a d Δ t − Δ R ~ i k ⋅ ( f ~ k − b i a ) ∧ ⋅ δ ϕ ⃗ i k ⋅ Δ t ] = ∑ k = i j − 2 [ Δ R ~ i k η k a d Δ t − Δ R ~ i k ⋅ ( f ~ k − b i a ) ∧ ⋅ δ ϕ ⃗ i k ⋅ Δ t ] … + Δ R ~ i j − 1 η j − 1 a d Δ t − Δ R ~ i j − 1 ⋅ ( f ~ j − 1 − b i a ) ∧ ⋅ δ ϕ ⃗ i j − 1 ⋅ Δ t = δ v i j − 1 + Δ R ~ i j − 1 η j − 1 a d Δ t − Δ R ~ i j − 1 ⋅ ( f ~ j − 1 − b i a ) ∧ ⋅ δ ϕ ⃗ i j − 1 ⋅ Δ t \begin{aligned} \delta \mathbf{v}_{i j}= & \sum_{k=i}^{j-1}\left[\Delta \tilde{\mathbf{R}}_{i k} \mathbf{\eta}_{k}^{a d} \Delta t-\Delta \tilde{\mathbf{R}}_{i k} \cdot\left(\tilde{\mathbf{f}}_{k}-\mathbf{b}_{i}^{a}\right)^{\wedge} \cdot \delta \vec{\phi}_{i k} \cdot \Delta t\right] \\ = & \sum_{k=i}^{j-2}\left[\Delta \tilde{\mathbf{R}}_{i k} \boldsymbol{\eta}_{k}^{a d} \Delta t-\Delta \tilde{\mathbf{R}}_{i k} \cdot\left(\tilde{\mathbf{f}}_{k}-\mathbf{b}_{i}^{a}\right)^{\wedge} \cdot \delta \vec{\phi}_{i k} \cdot \Delta t\right] \ldots \\ & +\Delta \tilde{\mathbf{R}}_{i j-1} \boldsymbol{\eta}_{j-1}^{a d} \Delta t-\Delta \tilde{\mathbf{R}}_{i j-1} \cdot\left(\tilde{\mathbf{f}}_{j-1}-\mathbf{b}_{i}^{a}\right)^{\wedge} \cdot \delta \vec{\phi}_{i j-1} \cdot \Delta t \\ = & \delta \mathbf{v}_{i j-1}+\Delta \tilde{\mathbf{R}}_{i j-1} \boldsymbol{\eta}_{j-1}^{a d} \Delta t-\Delta \tilde{\mathbf{R}}_{i j-1} \cdot\left(\tilde{\mathbf{f}}_{j-1}-\mathbf{b}_{i}^{a}\right)^{\wedge} \cdot \delta \vec{\phi}_{i j-1} \cdot \Delta t \end{aligned} δvij===k=i∑j−1[ΔR~ikηkadΔt−ΔR~ik⋅(f~k−bia)∧⋅δϕik⋅Δt]k=i∑j−2[ΔR~ikηkadΔt−ΔR~ik⋅(f~k−bia)∧⋅δϕik⋅Δt]…+ΔR~ij−1ηj−1adΔt−ΔR~ij−1⋅(f~j−1−bia)∧⋅δϕij−1⋅Δtδvij−1+ΔR~ij−1ηj−1adΔt−ΔR~ij−1⋅(f~j−1−bia)∧⋅δϕij−1⋅Δt -
δ
p
i
j
−
1
→
δ
p
i
j
\delta p_{ij-1} \to \delta p_{ij}
δpij−1→δpij
δ p i j = ∑ k = i j − 1 [ δ v i k Δ t − 1 2 Δ R ~ i k ⋅ ( f ~ k − b i a ) ∧ δ ϕ ⃗ i k Δ t 2 + 1 2 Δ R ~ i k η k a d Δ t 2 ] = δ p i j − 1 + δ v i j − 1 Δ t − 1 2 Δ R ~ i j − 1 ⋅ ( f ~ j − 1 − b i a ) ∧ δ ϕ ⃗ i j − 1 Δ t 2 + 1 2 Δ R ~ i j − 1 η j − 1 a d Δ t 2 \begin{aligned} \delta \mathbf{p}_{i j} & =\sum_{k=i}^{j-1}\left[\delta \mathbf{v}_{i k} \Delta t-\frac{1}{2} \Delta \tilde{\mathbf{R}}_{i k} \cdot\left(\tilde{\mathbf{f}}_{k}-\mathbf{b}_{i}^{a}\right)^{\wedge} \delta \vec{\phi}_{i k} \Delta t^{2}+\frac{1}{2} \Delta \tilde{\mathbf{R}}_{i k} \boldsymbol{\eta}_{k}^{a d} \Delta t^{2}\right] \\ & =\delta \mathbf{p}_{i j-1}+\delta \mathbf{v}_{i j-1} \Delta t-\frac{1}{2} \Delta \tilde{\mathbf{R}}_{i j-1} \cdot\left(\tilde{\mathbf{f}}_{j-1}-\mathbf{b}_{i}^{a}\right)^{\wedge} \delta \vec{\phi}_{i j-1} \Delta t^{2}+\frac{1}{2} \Delta \tilde{\mathbf{R}}_{i j-1} \boldsymbol{\eta}_{j-1}^{a d} \Delta t^{2} \end{aligned} δpij=k=i∑j−1[δvikΔt−21ΔR~ik⋅(f~k−bia)∧δϕikΔt2+21ΔR~ikηkadΔt2]=δpij−1+δvij−1Δt−21ΔR~ij−1⋅(f~j−1−bia)∧δϕij−1Δt2+21ΔR~ij−1ηj−1adΔt2
矩阵C:
χ i j = [ η i j T , γ i j T ] T ∼ N ( 0 15 × 1 , C i j ) η i j = [ δ ϕ i j T , δ v i j T , δ p i j T ] T ∼ N ( 0 9 × 1 , Σ i j η ) γ i j = [ δ b i j a T , δ b i j g T ] T ∼ N ( 0 6 × 1 , Σ i j γ ) \begin{aligned} \chi_{i j} & =\left[\eta_{i j}^{T}, \gamma_{i j}^{T}\right]^{T} \sim N\left(0_{15 \times 1}, C_{i j}\right) \\ \eta_{i j} & =\left[\delta \phi_{i j}^{T}, \delta v_{i j}^{T}, \delta p_{i j}^{T}\right]^{T} \sim N\left(0_{9 \times 1}, \Sigma_{i j}^{\eta}\right) \\ \gamma_{i j} & =\left[\delta b_{i j}^{a T}, \delta b_{i j}^{g T}\right]^{T} \sim N\left(0_{6 \times 1}, \Sigma_{i j}^{\gamma}\right) \end{aligned} χijηijγij=[ηijT,γijT]T∼N(015×1,Cij)=[δϕijT,δvijT,δpijT]T∼N(09×1,Σijη)=[δbijaT,δbijgT]T∼N(06×1,Σijγ)
根据 δ ϕ \delta \phi δϕ、 δ v \delta v δv、 δ p \delta p δp、 δ b a \delta b_{a} δba、 δ b g \delta b_{g} δbg,可以得到:
C i j = ( Σ i j η 0 9 × 6 0 6 × 9 Σ i j γ ) 15 × 15 C_{i j}=\left(\begin{array}{cc} \Sigma_{i j}^{\eta} & 0_{9 \times 6} \\ 0_{6 \times 9} & \Sigma_{i j}^{\gamma} \end{array}\right)_{15 \times 15} Cij=(Σijη06×909×6Σijγ)15×15
η
i
j
Δ
≜
[
δ
ϕ
⃗
i
j
T
δ
v
i
j
T
δ
p
i
j
T
]
T
\boldsymbol{\eta}_{i j}^{\Delta} \triangleq\left[\begin{array}{lll}\delta \vec{\phi}_{i j}^{T} & \delta \mathbf{v}_{i j}^{T} & \delta \mathbf{p}_{i j}^{T}\end{array}\right]^{T}
ηijΔ≜[δϕijTδvijTδpijT]T的递推形式如下
η
i
j
Δ
=
[
Δ
R
~
j
j
−
1
0
0
−
Δ
R
~
i
j
−
1
⋅
(
f
~
j
−
1
−
b
i
a
)
∧
Δ
t
I
0
−
1
2
Δ
R
~
i
j
−
1
⋅
(
f
~
j
−
1
−
b
i
a
)
∧
Δ
t
2
Δ
t
I
I
]
⏟
A
j
−
1
η
i
j
−
1
Δ
⋯
+
[
J
r
j
−
1
Δ
t
0
0
Δ
R
~
i
j
−
1
Δ
t
0
1
2
Δ
R
~
i
j
−
1
Δ
t
2
]
⏟
B
j
−
1
[
(
η
j
−
1
g
d
)
T
(
η
j
−
1
a
d
)
T
]
⏟
η
j
−
1
d
\begin{aligned} \boldsymbol{\eta}_{i j}^{\Delta}= &{\underbrace{ {\left[\begin{array}{ccc} \Delta \tilde{\mathbf{R}}_{j j-1} & \mathbf{0} & \mathbf{0} \\ -\Delta \tilde{\mathbf{R}}_{i j-1} \cdot\left(\tilde{\mathbf{f}}_{j-1}-\mathbf{b}_{i}^{a}\right)^{\wedge} \Delta t & \mathbf{I} & \mathbf{0} \\ -\frac{1}{2} \Delta \tilde{\mathbf{R}}_{i j-1} \cdot\left(\tilde{\mathbf{f}}_{j-1}-\mathbf{b}_{i}^{a}\right)^{\wedge} \Delta t^{2} & \Delta t \mathbf{I} & \mathbf{I} \end{array}\right]}} _{A_{j-1} }\boldsymbol{\eta}_{i j-1}^{\Delta} \cdots } \\ & +\underbrace{\left[\begin{array}{cc} \mathbf{J}_{r}^{j-1} \Delta t & \mathbf{0} \\ \mathbf{0} & \Delta \tilde{\mathbf{R}}_{i j-1} \Delta t \\ \mathbf{0} & \frac{1}{2} \Delta \tilde{\mathbf{R}}_{i j-1} \Delta t^{2} \end{array}\right]}_{B_{j-1}}\underbrace{ \begin{bmatrix} \left(\eta_{j-1}^{gd}\right)^T\\ \left(\eta_{j-1}^{ad}\right)^T \end{bmatrix} } _{\boldsymbol{\eta}_{j-1}^{d}} \end{aligned}
ηijΔ=Aj−1
ΔR~jj−1−ΔR~ij−1⋅(f~j−1−bia)∧Δt−21ΔR~ij−1⋅(f~j−1−bia)∧Δt20IΔtI00I
ηij−1Δ⋯+Bj−1
Jrj−1Δt000ΔR~ij−1Δt21ΔR~ij−1Δt2
ηj−1d
(ηj−1gd)T(ηj−1ad)T
Σ
i
j
η
\Sigma_{ij}^{\eta}
Σijη的递推形式如下:
Σ
i
j
η
=
A
j
−
1
Σ
i
j
−
1
η
A
j
−
1
T
+
B
j
−
1
Σ
η
B
j
−
1
T
\boldsymbol{\Sigma}_{i j}^{\eta}=\mathbf{A}_{j-1} \boldsymbol{\Sigma}_{i j-1}^{\eta} \mathbf{A}_{j-1}^{T}+\mathbf{B}_{j-1} \boldsymbol{\Sigma}_{\boldsymbol{\eta}} \mathbf{B}_{j-1}^{T}
Σijη=Aj−1Σij−1ηAj−1T+Bj−1ΣηBj−1T
其中:
Σ
η
\Sigma_{\eta}
Ση为IMU噪声的协方差矩阵
γ
i
j
=
[
δ
b
i
j
a
T
,
δ
b
i
j
g
T
]
T
\gamma_{i j} =\left[\delta b_{i j}^{a T}, \delta b_{i j}^{g T}\right]^{T}
γij=[δbijaT,δbijgT]T的递推形式如下
γ
i
j
=
γ
i
j
−
1
+
σ
j
−
1
d
\gamma_{i j}=\gamma_{i j-1}+\sigma_{j-1}^{d}
γij=γij−1+σj−1d
其中:
σ
k
d
=
[
(
σ
k
b
g
d
)
T
(
σ
k
b
a
d
)
T
]
T
\sigma_{k}^{d}=\left[\left(\sigma_{k}^{b g d}\right)^{T} \left(\sigma_{k}^{b a d}\right)^{T}\right]^{T}
σkd=[(σkbgd)T(σkbad)T]T
Σ
i
j
γ
\Sigma_{i j}^{\gamma}
Σijγ的递推形式如下:
Σ
i
j
γ
=
Σ
i
j
−
1
γ
+
Σ
γ
\begin{array}{l} \Sigma_{i j}^{\gamma}=\Sigma_{i j-1}^{\gamma}+\Sigma_{\gamma} \end{array}
Σijγ=Σij−1γ+Σγ