man Activit

ivity Recognition UsingSmartphones:

l n o :2 0 7

oduction

n e
70 a large growth in human life expectancy has been observed in
1 8 7 0

This growth has expanded in the whole world principally due to the
hievements in health care field. As a result, the proportion of elderly

splei
la is rapidly increasing. Aging people in general lives in isolated
ans, In addition to that some of them are not capable to live normally
pditions.

pke advantages trom health care facilities services. Building remote

itoring systems for elderly patients who live alone or without permanent

aking will improve their quality of life. For better decision making these
ote monitoring systems needs some regular and trustful information about
tients.

ject Origins:
has
tlfil remote monitoring systems' requirements Jorae uis Reves Ortiz
able to detect and
eloped a complete Human Activity Recognition Syste
in their daily living in
gnize different activities performed by humans
ine mode using smartphones. The recognition part of his system has already
ntested which is capable of predicting activities performed by users.
ESsary datasets of users' movements will be collected from smartphone
srs (accelerometer and gyroscope), processed and then fed to the prediction
del to recognize performed activities.

Ac and ayrc
signats
Machine Leaming
Atcelerometer Signal processing
model SVM Activities predictions
capable of
Features creation predicting the
Datasets generation
Gyroscope pata inputs activity pertormed Additional info such as:
|bythe user The user name
-Time
tphone Recognition
Data Processing Location.
Dched to the user
Mobiie application
Project Overview:
The goal of this project is to build a machine learning model in offline mod.
node
capable of processing signals collected using smart phone inertial sensors ar
producing useful datasets will be used as inputs of a machine learning model
capable of recognizing some of human daily activities (sitting., walking .)
included in the dataset (see datasets and Inputs section) with a low error
rate.
The signal processing pipeline and the final model could be used as a good
source of information about patient's daily activities needed by remote
monitoring systems mentioned earlier

The Experiment
The experiments were carried out with a group of 30 volunteers within an age
bracket of 19-48 years. They performed a protocol of activities
composed of six
basic activities: three static postures (standing,
sitting, lying) and three
dynamic activities (walking, walking downstairs and walking upstairs).
The
experiment also included postural transitions that occurred between the static
postures. These are: stand-to-sit, sit-to-stand, sit-to-lie, lie-to-
sit, stand-to-lie, and lie-to-stand. All the
participants were wearing a
smartphone (Samsung Galaxy S II) on the waist during the experiment
execution. They captured 3-axial linear acceleration and3-axial angular
velocity at a constant rate of 5OHz using the embedded accelerometer and
gyroscope of the device. The experiments were video-recorded to label the
data manually. The resulted dataset is stored and is
considered as the original
labelled dataset and from it different subsets would
be generated.
27thim:

Expenments

mo

eleration and gyroscope signals acquired:

Action performed by the users
Standing
Walking

Sitting Lying

ppadpondueioery

Conclusion:
From the processed signals presented in the picture above, they extract data
points and activity labels related only to the first six activities which
standing, sitting, lying, walking, walking downstairs and walking
are:

upstairs called Basic Activites(BA) to build the first

*The second version of this
version of this datas
datasets includes all the 12 activities
 Python programming for rapberry pi:
l n o :2 0 7

following:
this project we willneed the

Raspberry Pi
One LED
. One 330 Ohm resistor
Jumper wires
Breadboard

below:
snnect the components as shownin the wiring diagram

yvVV
limiting resistors should always
resistor. Current
S0 Ohm resistor is a current limiting LED is connected to a
GPIO pin
LEDs to the GPIO pins. If a n
ed when connecting which c a n damage the Raspberry
a resistor, the LED will
draw too much current,
ut
burn out the LED.

HON CODE:
 TugtRPi.GPIO as GPIO
import time
GPIO. setmode (GPIo. BOARD)
GPIO. setup (7, GPIO.OUT)
fo1 X in range ( 0 , 5 ) :

GPIO.output (7, True)

time.sleep (2)
GPIO.output (7, False)
time.sleep (2)
GPIO.cleanup (0

Explanation:

import RPi.GPIO as GPIO This imports the GPIO library for

Python
importtime- This imports the time library for Python
GPIO.setmodelGPIO.BOARD)
scheme the
This defines what
numbering
pins use. This is the simplest because it follows the
chart above.

GPIO.setup(7.GPIO.0OUT) -

This tells Python to use pin 7 where the

positive wire is plugged in
forxin rangel0,5): This sets the loop for how
want the LED to turn on many times you
and off. Right now it will
times. turn on and off 5

GPIO.output(7,True) -This tells Python to turn on pin 7

time.sleep(2)- This tells Python to leave pin 7 on for 2
seconas
GPIO.output(7, False)-This tells Python to turn off pin 7
time.sleep(2) -This tells Python to stay off for 2 seconds
GPIO.cleanup()- This just cleans everything up after it is nplete
co
 sion:from
this experiment we have learnt how to control
pncle
of led light by implementing python programming in
Output
e O u t p u t

Bpberry pi.
atroduction to Object Detection for Self Driving Cars:
Rollno:207

Introduction:

Today's world is full of data, and images form a significant part of

his data. However, to be put to any use, these images need to be
rocessed. Image Processing is thus the process of analyzing and
nanipulating a digital image primarily aimed at improving its
uality or extracting some information from it, which could then be

ISed.

ypical tasks in image processing include displaying images, basie

nanipulations like eropping, flipping, rotating, etc., Inmage
egmentation, Classification and feature extractions, Image
estoration, and Image recognition. Python becomes an apt choice
r such Image processing tasks. This is due to its growing
opularity as a scientific programming language and the free
ailability of many State of Art Image Processing tools in its
0system.
Color Features
The simplest feature we can get lrom mages consiste.o
values. For instanee, here is l raw ew
an image whie
containsaCar.

Building aClassifier:
A classic
approach is to first design a classifier
car
images trom non-car that
differentiate
can

an entire frame
images, and then run that classiler
sampling small patches
acu
that classified as
car
along the way. The paee
atehes
arethe desired
Work
properly, we detections. For this appro tu
must train our
non-car images. classifier to distingush cauand

To train
any classifier,
the two classes we
we need labeled data. cASC

would like to Lots ot t.

images. So we need distinguish are car andfranes d non-ear

samples of frames

available, we need to crop out both. If we only have suos

fived
regions and scale them n to
a a
 Car Not-car
Buthow exactly did we manage to create a car that could
Iriveitself2
imply hooking upa computer with some cameras to a car isn't
nough, because computers don't see images in the same way that
edo. Take a look:

s, we see elaborate shapes and edges that our big brains can
e together to make up a face. But to a computer, they see an
y of many numbers.
many
Convolutional Neural Networks CNNs):are used
Vision because of their amazing ability to inn Computer
understand snspacial TH
information. This means if I had a picture of a
rotate it. move it around, squeeze and
human
human,. even if CC
stretch it, the CNN
recognize that it is a human. would sa 11
L
Convolutional Neural Networks by themselves
image classification: given an image, are
mainly used in
assign a given class. For example, usingenetwork
a CNN
will aaccurate in
lesion dataset, it can then learn
to diagnose
trained on c a

from a given
image. different skin Can
cancers
But what IfI wanted to know more than
specifically, where exactly thejustobjectanisimage belonoc.
class, more if

Additionally, what if there are multiple located?

are different classes?
A objects within an image
5 cars and 20 self-driving
car cant
just know
that
people the area, it needs to know
are relative to
in that there are
itself in order to where they
navigate safely
This is where
object detection comes in. By
ntolutional Neural Network. we spicing up our
can its
repurpose amazing
lassification properties to also locate where the classes are in the
image. We can do so
through an
L0ok Once) which can performalgorithm
called YOLO (You Only
real-time object detection. perfect
for autonomous vehicles

onclusion
2essence, YOLO locates and classifies objects within an
age/video. Object Detection algorithms like YOLO. combined
th the many other sensors on a self-driving car like Li-Dar. allow
sto build
fully autonomous cars that can drive faster. sater. and
tter than any human can.