XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE
Motion Retargeting for Virtual Human
Shi Qu
Department of Space and Air Early
Warning
Air Force Early Warning Academy
Wuhan, China
qushi@nudt.edu.cn
Jianxun Liu
Department of Space and Air Early
Warning
Air Force Early Warning Academy
Wuhan, China
jianxunliu@sina.com
Tinxin Xu
Department of Space and Air Early
Warning
Air Force Early Warning Academy
Wuhan, China
tinxin@163.com
Liang Ma
Department of Space and Air Early
Warning
Air Force Early Warning Academy
Wuhan, China
maliang@163.com
Abstract—Present the concept of human lower limbs vector,
which can retain the primary feature of motion. After that,
present a method of human motion retargeting based on fixity
of lower limbs vector feature. This motion retargeting method
orients human lower limbs, which retargets the motion data
from original human skeleton model to target human skeleton
model which has different bone length. The retargeted motion
retains the primary feature of original motion. The results of
experiment show that our method has good motion retargeting
effect and high computing efficiency.
Keywords—virtual human, motion retargeting, motion
feature
I. INTRODUCTION
The method of human motion generation based on data-
driven has the advantages of easy realization and high
fidelity of generated motion. The key of human motion
generation based on data-driven is the acquisition of motion
data. The commonly used method is motion capture method,
which captures the motion information of performers
through the motion capture equipment. Motion data has a
strong pertinence. A given segment of motion data is often
bound to a human skeleton model with fixed topology and
bone length. When the motion data is used to drive other
human skeleton models, motion distortion will occur. This
feature of motion data restricts its wide application, because
it is impossible to capture the motion of all human bodies
with different bone lengths. In order to solve this problem,
people began to study the processing technology of motion
data, hoping to process the motion data of the original human
skeleton model, so that it can correctly drive the target
human skeleton model, and maintain the main motion
characteristics, which is motion redirection technology.
According to the different solutions, the implementation
methods of motion redirection can be divided into inverse
kinematics (IK) based methods, space-time constraints based
methods and physical constraints based methods. The
method based on IK is the most basic motion editing
technology. Its basic idea is to first determine the position
constraint of the end effector of the target bone model
according to the motion capture data and the relationship
between the target bone model and the original bone model,
and then the IK solver calculates the motion data of the
middle joint, so as to get the new motion. Gleicher et al. [1]
applied the spatiotemporal constraint method to motion
redirection. They designed a spatiotemporal constraint solver.
By optimizing the solution, a lot of calculation time was
avoided, the solution speed was improved, and the original
motion quality was maintained. The above motion
redirection methods usually only consider the geometric
position constraints of human motion, but not the physical
constraints, so they are not suitable for sports with high
dynamic performance, such as jumping, boxing, table tennis
and so on. Tak et al. [2] think that balance is an important
physical characteristic that makes human motion natural and
real, and propose a motion balance filtering method, they
think that balance constraint is an important physical
constraint. Pullen et al. [3] extended the scope of physical
constraints, including moment constraints and momentum
constraints in addition to balance constraints. They
transformed motion redirection into constraint state
estimation based on frame by frame Kalman filter. Seyoon
Tak et al. [4] further added dynamic and kinematic
constraints, and realized the physical based motion
redirection method using Kalman filter, and applied for the
US national patent for this method [5].
In this paper, a method of motion redirection for human
lower limbs is proposed, which keeps the main
characteristics of the retargeted backward motion through the
vectors of lower limbs, avoiding the motion distortion
phenomenon. At the same time, the efficiency of motion
redirection is high because there is no complex operation.
II. M
OTION DISTORTION
Because of the strong pertinence of the motion data,
when the same motion data is used to drive the human body
of different sizes directly, the motion distortion will occur.
For the lower limbs of human body, the common motion
distortion mainly includes suspension, ground penetration,
sliding, drift and motion deformation.
Figure 1 shows the penetration or suspension distortion
caused by directly driving different sizes of human body with
the same motion data. In the figure, (a) is normal motion, (b)
is the distortion of lower limbs penetrating the ground when
driving the human body model with longer bone length with
the motion data, (c) is the distortion of human body
suspension when driving the human body model with shorter
bone length with the motion data.