Motion synthesis and editing in low-dimensional spaces

Hyun Joon Shin, Jehee Lee
2006 Computer Animation and Virtual Worlds  
1 Human motion is difficult to create and manipulate because of the high dimensionality and spatiotemporal nature of human motion data. Recently, the use of large collections of captured motion data has added increased realism in character animation. In order to make the synthesis and analysis of motion data tractable, we present a lowdimensional motion space in which high-dimensional human motion can be effectively visualized, synthesized, edited, parameterized, and interpolated in both
more » ... and temporal domains. Our system allows users to create and edit the motion of animated characters in several ways: The user can sketch and edit a curve on low-dimensional motion space, directly manipulate the character's pose in three-dimensional object space, or specify key poses to create in-between motions. of the grid points (blue dots in Figure 7 ) and the shape of the curve is modified accordingly. Keyframing. The data-driven keyframing capability of our system was demonstrated with two sets of motion data: walk-and-pick and boxing. In the walk-and-pick example, we use a regular 20 × 20 grid (see Figure 8 (Top)). In the start pose, the character was swing the right leg in a walk cycle. In the end pose, the character was reaching the right hand forward. Our algorithm found a natural in-between motion in which the character took a short step forward, stopped, and reached the right hand forward. Our boxing data was about one minute long and included a variety of actions such as punching, dodging, ducking, and blocking punches. The boxing example required a denser 40×40 grid because the boxing data has a narrow passage in the valid region. Discussion Our approach scales relatively well with the size of motion data due to the local projection strategy. There is one issue to be addressed for achieving better scalability. Both MDS and Isomap requires O(n 2 ) memory space for maintaining the all-frames-to-frames dissimilarities, where n is the number of frames in the database. We observed that motion data tends to be partitioned into sets of coherently parameterizable motions and narrow connections between them. We envision the use of graph cut techniques to partition motion data 14 automatically. With this partitioning, the dissimilarity matrix could be stored compactly.
doi:10.1002/cav.125 fatcat:fwuazksynjcdvpyvyfaomchiaq