# Signal Processing Feature Analysis

Τεχνίτη Νοημοσύνη και Ρομποτική

24 Νοε 2013 (πριν από 4 χρόνια και 7 μήνες)

75 εμφανίσεις

Body Sensor Networks to Evaluate Standing
Balance: Interpreting Muscular Activities
Based on Intertial Sensors

Rohith Ramachandran

Lakshmish Ramanna

Roozbeh Jafari

Balakrishnan Prabhakaran

University of Texas at Dallas

Presented by,

Corey Nichols

Introduction

Why i
nterpret muscle activities for balance
performance based on intertial sensors
?

Rehabilitation, sports medicine, gait analysis, & fall
detection all can make use of a balance evaluation.

Inertial sensors currently in use, but do not measure
muscle activity directly

Measuring muscle activity may provide additional info

Goal

Investigate EMG signals to interpret standing balance

Use inertial sensors to help interpret these signals

Balance Parameters

[1] Mayagoitia, R.E., et al., Standing balance
evaluation using a triaxial accelerometer. Gait and
Posture, 2002. 16: p.55
-
59.

Parameters are classified as low, medium, and high

Want to analyze EMG signals to make the same
classifications using Linear Discriminant Analysis
(LDA)

LDA: Method in statistics and machine learning to find a
linear combination of features that best separates
multiple classes of objects or events
(source: wikipedia)

Evaluation Model

Uses the Balance Evaluation Model from [1]

Uses a single accelerometer

Height of the center of mass

Build and trace an acceleration vector

Building and tracing an
Acceleration vector

Building and tracing an
Acceleration vector

Combined Acceleration:

Directional angles using Cartesian Coordinates:

D is the combined coordinates in all three directions:

A
=
a
x
2
a
y
2
a
z
2
=
arccos
a
x
/
A
,
=
arccos
a
y
/
A
,
=
arccos
a
z
/
A
cos
=

d
z
/
D
,
d
x
=
Dcos
,
d
x
=
Dcos
Quantitative Features

Total Distance:

Mean Speed:

Mean Frequency:

Anterior/Posterior Displacement:

Medial/Lateral Displacement:

D
t
=

n
=
startpoint
endpoint
d
y
n

d
y
n
1
2
d
x
n

d
x
n
1
2
s
m
=
D
t
/
t
r
m
=
1
/
N

n
=
startpoint
endpoint
d
x
n
2
d
y
n
2
f
m
=
D
t
/
2
r
m
d
a
/
p
=
max

n
d
d
x
n

min

n
d
d
x
n
d
m
/
l
=
max

n
d
d
y
n

min

n
d
d
y
n
Quantitative Features

System Architecture

Inertial Sensor Subsystem

EMG Sensor Subsystem

Balance Platform

Inertial Sensor Subsystem

Body sensor network of two nodes

A tri
-
axial 2g accelerometer

Samples at 40Hz

Base station

Collects data over wireless

channel

Relays info to PC via USB

Sensor data is collected and

processed using MATLAB

EMG Sensor Subsystem

Four EMG sensors used

Measures electric activity

generated by muscle

contractions

Electrodes acquire EMG signal

Sample at 1000Hz

Signal is amplified and

band
-
pass filtered to 20
-
450Hz

Data is transferred to a PC and processed off line

Balance Platform

Balance ball (half sphere w/ standing platform)

Use a level to control

the experiment or for

coaching

Signal Processing Feature
Analysis

Five stages of operation

Data Collection

Parameter Extraction

Quantization

Feature Extraction on

EMG

Feature Analysis

Signal Processing Feature
Analysis

Data Collection

Accelerometer & EMG signals recorded every 4
seconds

Parameter Extraction

Extract 5 quantization factors using the accelerometer
data

Quantization

Classify data into 'low', 'medium' and 'high

Within 1
std. Dev. of the mean implies 'medium'

Signal Processing Feature
Analysis

Feature Extraction on EMG

Obtain an exhaustive set of statistical features from the
EMG signals

Signal Energy, Maximum Peak, Number of Peaks, Avg.
Peak Value, and Average Peak rate

Feature Analysis

Using LDA, extract significant features from EMG
signals

Determine if the EMG signals are representative of the
quantitative features for balance evaluation from the
accelerometer

Experimental Procedure

Subjects:

5 males aged 25
-
32 and 1.65
-
1.8m tall with no
disorders

Wore the accelerometer on a belt around the waist with
the sensor positioned in the back.

4 EMG electrodes attached on the lower leg

Right/Left
-
Front (Tibalis Anterior muscle)

Right/Left
-
Back leg (Gastrocnemius muscle)

Experimental Procedure

Sensors
:

Delsys “Trigger Module” allows the EMG to work
sychronously with the accelerometer

MATLAB tool sends the trigger

To EMG through the trigger module

To accelerometer through USB

MATLAB tool analyzes the data

Data was recorded every 4 seconds

Experimental Procedure

Test Conditions:

Nine test conditions

Two trials per condition

Experimental Results

90 trials performed

Classifies each trial into 'low',

'medium', & 'high' qualities

Done for each accelerometer

parameter

Each EMG feature is

assigned the same quality

label as its corresponding

accelerometer

data

Experimental Results

Made EMG signals representative of performance
parameter for balance evaluation

Used 50% of trials to find significant features

The remaining trials were for evaluation of the system

Extracted 5 signals from each of the four EMG

Form a 20 dimensional space that is representative of
some muscle activity properties

LDA is used to select the most prominent feature from
the subset

Experimental Results

Uses the k
-
Nearest Neighbor classifier to determine
the effectiveness of the EMG features

K
-
NN classifies objects using

training examples

Questions?

Related Work

A lot of work has been done based on human
performance and quality of balance

A study on children compared EMG with
kinetic parameters for balance responses
shows that muscle activities contribute to
balance

This is the first work that uses inertial sensors
to help interpret EMG signals

Conclusion & Future Work

Uses acceleration and muscle activity data to
perform an analysis during standing balance

Break the accelerometer data down into five metrics

Prominent features are extracted from EMG signals
using the accelerometer data to evaluate the balance

Future goals:

Integrate a “gold standard balance system” with their
experiments

deploying a system that performs the data processing in
real
-
time