Rotation and Orientation: Affine Combination Jehee Lee Seoul National University
Applications What do we do with quaternions ? –Curve construction Keyframe animation
Applications What do we do with quaternions ? –Filtering Convolution
Applications What do we do with quaternions ? –Statistical analysis Mean
Applications What do we do with quaternions ? –Curve construction Keyframe animation –Filtering Convolution –Statistical analysis Mean It’s all about weighted sum !
Weighted Sum How to generalize slerp for n-points –Affine combination of n-points Methods –Re-normalization –Multi-linear –Global linearization –Functional Optimization
Inherent problem Weighted sum may have multiple solutions –Spherical structure –Antipodal equivalence
Re-normalization Expect result to be on the sphere –Weighed sum in R –Project onto the sphere 4
Re-normalization Pros –Simple –Efficient Cons –Linear precision –Singularity: The weighted sum may be zero
Multi-Linear Method Evaluate n-point weighted sum as a series of slerps Slerp
Multi-Linear Method Evaluate n-point weighted sum as a series of slerps Slerp
De Casteljau Algorithm A procedure for evaluating a point on a Bezier curve t : 1-t P(t)
Quaternion Bezier Curve Multi-linear construction –Replace linear interpolation by slerp –Shoemake (1985)
Quaternion Bezier Spline Find a smooth quaternion Bezier spline that interpolates given unit quaternions –Catmull-Rom’s derivative estimation
Quaternion Bezier Spline Find a smooth quaternion Bezier spline that interpolates given unit quaternions –Catmull-Rom’s derivative estimation
Quaternion Bezier Spline Find a smooth quaternion Bezier spline that interpolates given unit quaternions –Catmull-Rom’s derivative estimation –Bezier control points (q i, a i, b i, q i+1 ) of i-th curve segment
Multi-Linear Method Slerp is not associative
Multi-Linear Method Pros –Simple, intuitive –Inherit good properties of slerp Cons –Need ordering Eg) De Casteljau algorithm –Algebraically complicated
Global Linearization
Pros –Easy to implement –Versatile Cons –Depends on the choice of the reference frame –Singularity near the antipole
Functional Optimization In vector spaces –We assume that this weighted sum was derived from a certain energy function
Functional Optimization In vector spaces Functional Minimize Weighted sum
Functional Optimization In orientation space –Buss and Fillmore (2001) Spherical distance Affine combination satisfies
Functional Optimization Pros –Theoretically rigorous –Correct (?) Cons –Need numerical iterations (Newton-Rapson) –Slow
Summary Re-normalization –Practically useful for some applications Multi-linear method –Slerp ordering Global linearization –Well defined reference frame Functional optimization –Rigorous, correct
Summary We don’t have an ultimate solution An appropriate solution may be determined by application More specific problems may have better solutions –For convolution filters, points have an ordering