Você está na página 1de 12

SMART HOME

TECHNOLOGY FOR
ELDERLY PEOPLE USING
KINECT SENSOR WITH
THE HELP OF ANN
By: Arju Utpal
Abhishek Kumar
Faizanul Haq
Devashrut

ABSTRACT:

The control of elderly people has always been a


major concern in hospitals as in homes.
Our aim is to control and surround elderly people
with comfort, safety in their homes while respecting
their privacy.
We propose a monitoring system based on Kinect
sensor to control and monitor elderly people with
less mobility through hand gestures .

SYSTEM OVERVIEW

Gesture recognition using Kinect sensor

Kinect provides with the position (X, Y, Z) of the user joints 30 frames per second. It
calculates the co-ordinates of the patient position and using
sensor it will track the focus gesture whenever elderly people moves their body parts.

PROJECT CHALLENGES
MATLAB Kinect interfacing
Acquiring and understanding of co-ordinates joints
Features decision and extraction
ANN training and testing
Interfacing of software and hardware to make complete system

ARTIFICIAL NEURAL NETWORKS (ANN)

ARTIFICIAL NEURAL NETWORK IS INSPIRED


BY BIOLOGICAL NEURON MODEL.
IT IS PRESENTED AS SYSTEM OF
INTERCONNECTED NEURONS WHICH
EXCHANGE MESSAGES BETWEEN THEM.
ARTIFICIAL NEURON IS A MATHEMATICAL
MODEL CONCEIVED AS A MODEL OF
BIOLOGICAL NEURONS.

LAYERS OF ANN
The basic idea is that you have three layers of "nodes."

The nodes have values of 0.0 to 1.0, where 0 represents fully


inactive "off" and 1 represents fully active "on" with many
values in between.

The three layers are an input layer, an output layer, and a


"hidden" layer in the middle (hidden means neither input nor
output, so not exposed to the outside world).

The nodes are linked by connections which have a "weight"


("w" in the figure) that are analogous to synapses in the
brain.

Signal values propagate from the inputs, through the


connection weights to the hidden nodes, and then onward
through more connection weights to the output nodes.

HOW ANN WORKS

In the given picture,the input to the given node will be


a(0)*w(0,i)+a(1)*w(1,i)+a(2)*w(2,i)+.. +a(j)*w(j,i)

This input function will be fed into the activation


function(which is sigmoid function) in our case.

After passing the given input function through the


activation function,the output of the given node will be
calculated.

Here,

a(0)- first input node

a(j)- (j+1)th input node

W(0,i)- connection weight between first input node and


the given node.

BACKPROPOGATION
We use backpropogation mechanism to train the ANN.

At first we give random connection weights between the nodes.

We will get some error in the output as weights which we have


assigned are random weights.

We will compare the output generated with the expected output.

We will adjust the connection weights between the hidden node and
output node to make the output more closer to the expected output.

ANN is a slow learner and we will have to repeat this process until we
get the correct output which we expected.

ANN TRAINING
INPUT

F.E

INPUT

F. E
1

INPUT

INPUT

F. E

F. E

WEIGH
T

ANN TESTING

TRAINED
WEIGHT

INPUT

F.E

WEIGH
T

RESUL
T

FEATURES EXTRACTION
KINECT

EXTRACTION OF
COORDINATES OF
JOINTS

MICROCONTROLLER

RF Rx
EXTRACT
FEATURES

ANN

RF Tx

GET
RESPECTIVE
ACTION

PERFORM RESPECTIVE
ACTION

TRANSMIT ACTION
SIGNAL TO
MICROCONTROLLE
R VIA RS232

RELAY

CONCLUSION
To continuously control the health of elderly people, it is
necessary to monitor them especially from afar. Taking care for
people with reduced mobility is a part of medicine still in
development. An automatic monitoring system seems to be
interesting in the medical side.

Você também pode gostar