原文:
Visual servo control. I. Basic approaches. Chaumette(2006)
Visual servo control. II. Advanced approaches. Chaumette(2007)
Basic formulas
s(m(t),a)
Compute a vector of k visual feature, with image coordinates of interest pointsm(t) and additional knowledge about the system a (eg. camera intrinsic matrix).e(t)=s(m(t),a)−s∗
The aim of all vision-based control schemes is to minimize an error e(t) . s∗ , the desired values of the features, assumed to be constant.vc=(vTc,wTc)T
vc , the instantaneous linear velocity of the origin of the camera frame.
wc , the instantaneous angular velocity of the camera frame.s.=Lvc
L∈Rk×6 , unknown, named the interaction matrix or the feature Jacobian related to s .e.=Lvc
Derived from (2)(4).vc=−λL+e
Come true if we assume that e.=−λe , which means the error has an exponential decrease.
L+∈R6×k , the Moore-Penrose pseudo-inverse.
If L is row full rank,L+=(L+L)−1LT .
If L is column full rank,((LT)+)T .
If L is not full rank, take SVD :L=UDV . Then, L+=VSUT , S(i,i)=⎧⎩⎨01D(i,i)D(i,i)=0D(i,i)!=0
Questions
- How should s be chosen?
- What is the form of
L ? - How should we estimate L+^ ?
- What are the performance characteristics of the resulting closed-loop system?
Classical Image-Based Visual Servo
Basic definition
m=(u,v) , the coordinates of the interest points in pixel units.
a=(cu,cv,f,α) , the set of camera intrinsic parameters: cu and cv are the coordinates of the principal point, f is the focal length,α is the ratio of the pixel dimensions.
s=x=(x,y) , calculate from the 3D point (X,Y,Z) , { xy=X/Z=(u−cu)/fα=Y/Z=(v−cv)/fL
Relation between 3D point variation and pixel point variation