[其它] PROGRESSIVE TRANSITIONS USING BODY PART MOTION GRAPHS

[复制链接]

彬彬

797 主题	1 听众	1万积分

资深设计师

Rank: 7 Rank: 7 Rank: 7

纳金币: 5568
精华: 0

电梯直达

楼主

发表于 2011-12-28 09:09:43 |只看该作者 |倒序浏览

Abstract

In this work we describe a preliminary method for progressive transitions

in human locomotions. To achieve this, motion graphs have

been used to synthesize body part transitions and every part has

been synchronized with the other parts using time scaling. Lastly,

we have compared transition costs and graph connectiviy of our

method to standard motion graphs. The results are promising to allow

better smoothness and response in motion clips concatenation.

1 Introduction

Automatic concatenation from motion capture clips in order to create

a larger stream of movements has been a wide research area in

3D character animation. In 2002, motion graph appears in various

works such as [Kovar et al. 2002][Lee et al. 2002]. Motion graph

consists in the assembling of motion capture data in a graph connecting

similar frames that are below a similarity pose threshold.

Distance metrics are computed to know this similarity. Common

distance metrics compute some function to compare frames, taking

into account full body posture. To refine this comparison, weights

are assigned to joints giving more or less importance to them. These

weights are of great importance and can change the transition points

between motion clips [Wang and Bodenheimer 2008]. Finally, motions

are connected satisfying user input.

Searching similar full body postures, as in standard motion graphs

(SMG), can sometimes be tricky. Two frames having the same posture,

except for some joints, could overcome the threshold similarity

even though the whole body is similar enough. Not being

able to find similar positions affects graph connectivity and reduces

the possibility of transitions between different kind of motions. In

Fig. 1 there are some transition matrices (distance between frames)

clustering joints in body parts (BP). As we can see, good transition

points (dark zones) are located in close positions of each map, but

not exactly in the same place. These differences show that transitioning

all joints at same time (in a SMG way) could be not optimal.

Therefore, we propose to divide the body into body parts and transition

each part independently. Some other works use similar body

part segmentation [Jang et al. 2008] [Ng et al. 2010], but they focus

on attaching body part motions from different motions in order to

enlarge a motion capture database. However, this is not our goal,

we are focused on searching good transition points for every body

part and generating their specific transitions. The main problem

with this approach is synchronization. Locomotions are the sum of

body part movements with a specific synchrony and cadence. We

deal with this issue searching transitions for each part of the body in

a time window, and then we edit submotions to ensure synchronization

and coherence. This process will be properly described in next

section. We have called our method progressive transitions using

body part motion graphs (BPMG’s).