Transcript Image-Based Health Measurements Advisors

Image-Based Health Measurements
Previous Research
System Details
Issues
Unfiltered
Signal
Bandpass
Filter (FL, FH)
Filtered Signal
Bandpass Filter
• Filter produces
output signal only
between low and
high frequency
specified
• Used in program to
filter out
unnecessary
components of
signal
Amplitude
Time
Figure 1. Moving average of
distorted sine wave
•
•
•
Difficulty capturing the
region of interest in
recorded videos
Accuracy of calculated
measurements
Accounting for shadows
and movements in recorded
videos
Future goals
•
•
•
•
Amplitude
Picard 2010
Moving
Average
Distorted
• Volumetric changes in
Signal
• Modify input signal
facial blood vessels
to remove
modify the path length
Irregularities
of ambient light
•
Output
is
“smooth”
Taking Moving
Previous research uses
version of input
Average
this detect
• Used in program to
cardiovascular events
eliminate high
• The region of interest is
frequencies in signal
Smoothened
decomposed into RGB
Signal
to help with target
channels and spatially
signal detection
averaged
• Heart rate is calculated
Example Outputs
from a frequencydomain signal
Picard 2011
• Extend method in
Picard 2010 to quantify
multiple physiological
parameters (HR, RR,
HRV)
• Obtain Interbeat
Intervals (IBI) from
time-domain signals
• Use mean IBI to
calculate heart rate
Advisors: Prof. Delp
Prof. Schwichtenberg
Graduate Mentor: Jeehyun Choe
Members: Junjie Wang
Pradeep Lam
Increase accuracy of
calculated measurements
Convert program from
Matlab to OpenCV
Refine respiration rate code
Increase movability of
subject
References
Frequency
Figure 2. Sine wave run through
bandpass filter
M Poh, D McDuff, R Picard. "Non-contact,
automated cardiac pulse
measurements using video imaging and blind
source separation." Optical Society of America,
2010.
M Poh, D McDuff, R Picard. "Advancements
in Noncontact, Multiparameter
Physiological Measurements Using a
Webcam." IEEE Transactions on Biomedical
Engineering, vol.58, Jan 2011.