Transcript 下載/瀏覽
3D Human Pose Recognition for
Home Monitoring of Elderly
Bart Jansen, Frederik Temmermans and Rudi Deklerck
Proceedings of the 29th Annual International
Conference of the IEEE EMBS
Cité Internationale, Lyon, France August 23-26, 2007.
Presenter: Yi-Shing Chen
Professor: Dr. Yen-Ting Chen
Date: 2009/12/22
1
Outline
Introduction
Method
● 3D Camera
● Pose recognition algorithm
Experiment and Results
Discussion and Applications
Ethical Issues
Conclusion
Reference
2
Introduction
The old population has increased
dramatically.
The growing need for beds in nursing homes
causes governments to stimulate elderly to
live longer in their natural home environment.
3
Introduction
Falls are the most common cause.
Average 33% of the seniors will fall in a year.
Falls also result in a decrease of life quality.
4
Introduction
The elderly become so afraid of falling that
they limit their activities.
Falls will remain an important cause of the
loss of independence of elderly.
5
Introduction
Many technical aids have been developed in
the prevention and detection of falls.
A
fall detector is a small portable device.
Detection
results are typically very good, on
condition that the device is worn correctly by
the elderly.
6
Introduction
3D camera is used for performing visual fall
detection, this approach is new.
A framework for the monitoring of elderly is
introduced.
► Standing
► Sitting
► Lying down
7
Method-3D Camera
3D Camera
The
camera is an active device, emitting
modulated infrared light.
Depth
It’s
information is provided.
based on the time-of-flight (TOF) principle.
Time-of-Flight :時差測距 或又稱飛時測距
8
Method-3D Camera
Time-of-Flight (TOF)
3D
laser scanner is an active scanner.
The rangefinder finds the distance of a surface
by timing the round-trip time.
9
[引用:維基百科http://zh.wikipedia.org/zh-tw/%E4%B8%89%E7%B6%AD%E6%8E%83%E6%8F%8F%E5%84%80]
Method-3D Camera
The depth information is readily available, without
heavy calculations.
3D cameras do not provide inaccurate depth
information in regions with poor texture information.
The image resolution is rather low.
Once 3D cameras will become more widely used, the
price is expected to drop significantly.
10
Method-Pose recognition algorithm
The pose recognition algorithm consists of three
steps.
First step
The human silhouette is extracted form the gray level
image by subtracting the image from the background.
The background image is calculated using a running
average over many frames.
Image process
Thresholding
Smoothing
Erosion
Dilation
11
Method-Pose recognition algorithm
Second step
The
center of the silhouette is calculated by
fitting an ellipse to the blob .
By thresholding on the width/height of the
ellipse, blobs not related to human silhouettes
can be discarded.
12
Blob identified as person
Method-Pose recognition algorithm
Third step
The 3D position of the silhouette’s center is
calculated in a defined coordinate system.
Silhouette’s
center needs to be transformed into
room coordinates.
The
height above the ground of the silhouette is
thresholded.
Standing
Sitting (z≦75)
Lying (z≦ 35)
13
Method-Pose recognition algorithm
近
已知道的高度
利用time of flight算出來的深度
75 cm
35 cm
遠
Experiment and Results
The 3D camera was positioned in the corner
of a living room.
Conditions
A
male subject of 27 years old
Different camera positions
Different light conditions
Different clothing
A week
15
Experiment and Results
m
Standing
70cm
Sitting
35cm
Lying
frame
16
Experiment and Results
The plot shows that lying down, sitting and
standing can clearly be distinguished.
walking sequences
sitting sequences
lying sequences
17
Discussion and Applications
The monitoring approach proposed here is
targeted to the monitoring of elderly.
In the future activity analysis framework will
be validated in the hospital.
The authors will investigate the trajectories
defined by the 3D position of the monitored
subject’s center.
18
Discussion and Applications
This information could let us know a lot way
of 3D pose.
To overcome the wrong judgment, like bend
down.
19
Ethical Issues
Right of privacy
The
important issues are the storage and
access of the monitored data.
Architecture of two components
At
the client side, there is the camera, together
with a processing unit.
At the server side, the monitored data is
stored in the electronic patient record.
20
Ethical Issues
Solution
At
the client are running, no images are
transmitted from the client to the server.
Only
the calculated activity information (e.g.
the position of the patient) is transmitted.
21
Ethical Issues
It’s guaranteed that the captured 3D images
are never stored, nor at the client or the
server side and that they are never
transmitted over the Internet.
client
server
沒
有
影
像
傳
送
22
Position information data
Ethical Issues
It is indeed true that patients reveal some
aspects of their privacy.
It is a choice to reveal information about their
behavior for health.
They could be able to quit the program at all
time.
23
Conclusion
This paper presents a simple but reliable 3D
pose recognition algorithm applied on images.
Author described the reliability of the pose
classification into the classes standing, sitting,
and lying down.
The information used to derive various
features which correlate with the well being of
the elderly.
24
Reference
Bart Jansen, Frederik Temmermans and Rudi Deklerck,3D
human pose recognition for home monitoring of elderly,
Proceedings of the 29th Annual International, Conference of the
IEEE EMBS, Cité Internationale, Lyon, France, August 23-26,
2007.
維基百科-三維掃瞄儀
http://zh.wikipedia.org/zhtw/Wikipedia:%E9%A6%96%E9%A1%
B5
25
Thank you for your attention
26