Animation From Observation: Motion Editing

1. Animation From Observation: Motion Editing • Dan Kong • CMPS 260 Final Project

2. Why we need motion editing • Reuse motion capture data to different character and different actions • Create infeasible motion which is hard to get by using motion capture • Change of intent which can not be predicted before the motion caputre

3. In this project, I will implement a prototype motion editing system that has the following features • Multiresolution motion filtering • Multitarget motion interpolation • Motion path editing

4. What is Motion Multiresolution filtering • Basically, it is a technique that originally used in image processing now applied to the motion parameters of an articulated figure. • Intuition of this method: low frequencies contain gross and smooth motion while high frequencies contain detail, abrupt motion and most of the noise.

5. Motion filtering algorithm • The number of frames determine how many frequency bands (fb) for each signal: let 2n < m < 2n+1, then fb = n. • Calculate the low-pass sequence of all fb signal by convolving the signal with a low-pass filter, obtaining Gk, Gk-1...G0 where Gfb is the DC component of the signal (0=<k<=fb) • Obtain the band- pass filter bands: L k = Gk – Gk+1

6. Motion filtering algorithm (continued) • Adjust gains for each band and multiply Lk by their current gain values. • Reconstruct the motion signal: G0 = Gfb +

7. Multitarget motion interpolation • A process widely used in computer animation to blend between different models. • Facial animation: By blending the corresponding parameters of different face, we can control the expression. • A happy walk + a sad walk = ?

8. Multitarget motion interpolation (continued) • Mix multitarget interpolation and multiresolution filtering to blend the frequency bands of two or more motion separately. • Need to establish the time correspondence between two signals.

9. Motion path editing • What is path editing: • Altering previously motion capture data to follow a different path • Why Path Editing: - Using original motion in a new environment - Dynamic application: walking to a goal location

10. Alternative Approach • Capture all desired motions • Disadvantages of this approach: - Can not possible predict every possible required motion - Such a motion library are expensive to create - Unwieldy to control as it expands

11. Algorithm Overview • Basic Idea: Represent the motion relative to the path. As the path is altered, the orientation of character updates accordingly. • Algorithm - Automatic path generation - Calculate initial path coordinate system - Path Editing - Computing the changed path coordinate system

12. Path Description • Why B-Spline • Local control without affecting global shape • continuous • Path Coordinate Representation - p(t) is the coordinate center. The corresponding transformation matrix is P(t) - R(t) is the orientation of the coordinate system. - P(t)R(t) is the path’s coordinate system - R(t)-1P(t)-1 is the transformation from global to path’s local coordinate system

13. Initial Path Generation • Least-Square Fitting • Using the root translation as the fitting point. • Compute average Y translation to constrain the path to lie in X-Z plane. • Store the R0(t)-1P0(t)-1 for later editing

14. Path Editing • After Altering the curve • P(t)R(t) R0(t)-1P0(t)-1 is computed as the coordinate system for the new path • Once the P(t) is updated, the corresponding R(t) must be computed accordingly

15. Future Work • Generate more frames when the path is streched • Add constraint to avoid geometric violation.

16. References “Motion Path Editing” “Motion Signal Processing” “Motion Warping” “Multiresolution Curves” “Hierarchical B-Spline refinement”

17. Thanks!!!