
data for machine learning. The lack of physical exertion
and absence of motion makes this class of human activities
amenable to relatively simple biomechanical models similar
to the ragdoll models used in video games [
35].
We apply this insight to the problem of using a pressure
image to estimate the 3D human pose and shape of a per-
son resting in bed. This capability would be useful for a
variety of healthcare applications such as bed sore manage-
ment [
17], tomographic patient imaging [18], sleep studies
[
9], patient monitoring [10], and assistive robotics [13]. To
this end, we present the PressurePose dataset, a large-scale
synthetic dataset consisting of 3D human body poses and
shapes with pressure images (Fig. 1, left). We also present
PressureNet, a deep learning model that estimates 3D hu-
man body pose and shape from a low-resolution pressure
image (Fig. 1, right).
Prior work on the problem of human pose estimation
from pressure images [
9, 13, 18, 22, 29] has primarily used
real data that is challenging to collect. Our PressurePose
dataset has an unprecedented diversity of body shapes, joint
angles, and postures with more thorough and precise anno-
tations than previous datasets (Table
2). While recent prior
work has estimated 3D human pose from pressure images,
[
9, 13], to the best of our knowledge PressureNet is the first
system to also estimate 3D body shape.
Our synthetic data generation method first generates di-
verse samples from an 85 dimensional human pose and
6215
0
静止的身体:使用合成数据从压力图像中估计3D人体姿
势和形状
0
HenryM.Clever1,ZackoryErickson1,ArielKapusta1,GregTurk1,C.KarenLiu2和CharlesC.Kemp1
0
1佐治亚理工学院,美国亚特兰大,2斯坦福大学,美国斯坦福
0
{henryclever,zackory,akapusta}@gatech.edu,turk@cc.gatech.edu,karenliu@cs.stanford.edu,charlie.kemp@bme.gatech.edu
0
图1.
左:PressurePose数据集包含206K个3D人体姿势和形状,压力图像是通过物理模拟生成的,模拟了关节刚体模型和软体模型
在床和压力感应垫上的情况。右:PressureNet是一个在合成数据上训练的深度学习模型,在真实数据上表现良好:输入压力
图像和性别(in),输出3D人体网格(out),用于参考的RGB图像(ref)。
0
摘要
0
人们在床上休息的时间很长。为这种活动进行3D人体姿势和
形状估计将有许多有益的应用,但是由于被床上物品的遮挡
,直线视觉感知变得复杂。压力感应垫是一个有前途的替代
方法,但是收集大规模的训练数据具有挑战性。我们描述了
一种基于物理的方法,该方法模拟了人体在床上休息时的情
况,并提供了Pressure-Pose,一个具有206K压力图像、3
D人体姿势和形状的合成数据集。我们还提出了PressureNe
t,这是一个深度学习模型,可以根据压力图像和性别估计
人体姿势和形状。PressureNet包含一个压力图重建(PMR
)网络,用于模拟压力图像的生成,以促进估计的3D人体模
型和压力图像输入之间的一致性。在我们的评估中,Pressu
reNet在各种姿势的真实数据中表现良好,尽管它只是使用
合成数据进行训练。当我们去除PMR网络时,性能显著下降
。
0
1.引言
0
人类在休息时会选择可以轻松维持的姿势。我们的主要观点
是,人体在休息时可以被建模得足够好,以生成合成的