BlanketSet - A clinical real-world in-bed action recognition and qualitative semi-synchronised MoCap dataset

2025-04-27 0 0 1.2MB 4 页 10玖币

侵权投诉

BlanketSet - A clinical real-world in-bed action

recognition and qualitative semi-synchronised

MoCap dataset

Jo˜

ao Carmona∗†, Tam´

as Kar´

acsony∗†‡, Jo˜

ao Paulo Silva Cunha∗† Senior Member, IEEE

∗Center for Biomedical Engineering Research, INESC TEC, Porto, Portugal

†Faculty of Engineering (FEUP), University of Porto, Porto, Portugal

‡Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA

Abstract—Clinical in-bed video-based human motion analysis

is a very relevant computer vision topic for several relevant

biomedical applications. Nevertheless, the main public large

datasets (e.g. ImageNet or 3DPW) used for deep learning

approaches lack annotated examples for these clinical scenarios.

To address this issue, we introduce BlanketSet, an RGB-IR-

D action recognition dataset of sequences performed in a

hospital bed. This dataset has the potential to help bridge the

improvements attained in more general large datasets to these

clinical scenarios. Information on how to access the dataset is

available at rdm.inesctec.pt/dataset/nis-2022-004.

Index Terms—Action recognition dataset, RGB-IR-D Video,

Motion Capture, Human Pose Estimation, Epilepsy monitoring,

Sleep analysis

I. INTRODUCTION

Human motion analysis in videos has recently seen drastic

improvements due to advances in Deep Learning (DL).

However, almost all the research in this area was focused on

the most general context of people standing up without any

occlusions. This, paired with the difﬁculty of generalizing

DL models beyond the domains of their training datasets,

means that human motion analysis for patients in bed is

sorely lacking when compared with more common scenarios.

In order to bridge this domain gap, we introduce Blan-

ketSet, an action recognition dataset of people in a hospital

bed which consists of 405 RGB-IR-D recordings of 14

participants performing 8 different movement sequences.

Each recording was repeated 3 times with different levels

of blanket occlusion, so it can be used not just for action

recognition, but also for qualitative evaluation of human

motion analysis systems. Even though there are no ground

truth poses or body shapes, the actions are carried out in

a semi-synchronized manner, as described in (III-C). As an

example of this second use case, in this work we used

BlanketSet to evaluate the real-world performance of the

pipeline implemented in [1] and found that BlanketGen

improved the performance of a DL human pose estimation

system with statistical signiﬁcance when full-body blanket

occlusions were present in the real world.

This work is ﬁnanced by National Funds through the Portuguese funding

agency, FCT - Fundac¸˜

ao para a Ciˆ

encia e a Tecnologia, within project

LA/P/0063/2020 as well as under the scope of the CMU Portugal (Ref

PRT/BD/152202/2021).

II. RELATED WORK

In-bed movement monitoring is done in different contexts,

the most relevant ones being epileptic seizure analysis and

sleep analysis.

In the area of automatic epileptic seizure classiﬁcation,

there has been signiﬁcant research, in [2] and [3] a system

was set up to record RGB-IR-D videos synchronized with

electroencephalography data of patients staying at the Uni-

versity of Munich Epilepsy Monitoring Unit (EMU) and then

have them manually labeled by clinicians. IR data acquired

with that system was used in [4] and [5] to explore deep

learning action recognition at EMUs. Data recorded with this

system is very promising for future research in this area but

is not publicly available, which limits its accessibility.

In the area of sleep analysis, the use of data collected

from cameras positioned above the bed has been explored

as a low-cost non-invasive approach with promising results

[6]–[8]. However, the only image dataset of people lying in

beds that is publicly available is SLP [9], which contains

images of people in static positions along with 2D position

ground truth; it is an useful resource but lacks the temporal

dimension that is very relevant in contexts with considerable

movement such as the analysis of epileptic seizures.

Image and video-based DL systems have been shown to

work exceptionally well in tasks related to human motion

recognition and classiﬁcation in general use cases [10]–[12],

therefore it is reasonable to expect that they would also be

able to perform well in the more speciﬁc use cases of sleep

analysis and epileptic seizure analysis. As with everything

related to deep learning, however, large amounts of data are

required and, due to the lack of publicly available datasets

in these areas, each separate research effort has also had to

include the acquisition of its own dataset which signiﬁcantly

hampers the efﬁciency of the research. The acquisition and

publication of datasets such as [13]–[17] were instrumental to

the incredible results attained in the more general use cases.

There have also been research efforts into utilizing these

broader datasets in epileptic seizure analysis: [4] employed

transfer learning from the Kinetics-400 [16] and ImageNet

[17] datasets to discriminate between two classes of epileptic

seizures, and BlanketGen [1] augmented the 3DPW dataset

arXiv:2210.03600v3 [cs.CV] 19 Mar 2023

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

BlanketSet-Aclinicalreal-worldin-bedactionrecognitionandqualitativesemi-synchronisedMoCapdatasetJoaoCarmonay,Tam´asKar´acsonyyz,JoaoPauloSilvaCunhaySeniorMember,IEEECenterforBiomedicalEngineeringResearch,INESCTEC,Porto,PortugalyFacultyofEngineering(FEUP),UniversityofPorto,Porto,PortugalzRoboti...

展开>> 收起<<

BlanketSet - A clinical real-world in-bed action recognition and qualitative semi-synchronised MoCap dataset.pdf

共4页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

BlanketSet - A clinical real-world in-bed action recognition and qualitative semi-synchronised MoCap dataset

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: