Hough

Object tracking

Work team:
Omar Al-aklouk
Hanan akelan
Heyem awad

Under the supervision D: Ahmed abo abs

2019/2020
Table of Contents:
-introduction
-problem statement
-research question
-objective
-scope
-significance of study
-literature review
-references

Introduction:
Research on how Hough works that have ability for image recognition through python in our
research we will speak about object tracking and how it works.
It is by looking and seeing we come to know the world we live in. The environment that

surrounds us is filled with endless types of objects and impressions. Vision is in other words
a means to gain an understanding for the world around us. Exactly how the visual system

works remains a mystery to be solved, even though physiologists have been investigating the

phenomena for decades. When speaking about vision, by replacing the living creature with

a computational instrument, we have the broad and abstract expression computer vision.

It can be summarized as the process of computers analysing digital images or videos and

gaining a high level understanding from it.

Object tracking is an area within computer vision which has many practical applications

such as video surveillance, human-computer interaction, and robot navigation [19]. It is

a well-studied problem, and in many cases a complex problem to solve. The problem of

object tracking in video can be summarized as the task of finding the position of an object

in every frame. The ability to track an object in a video depend on multiple factors,

like knowledge about the target object, type of parameters being tracked and type of video

showing the object.

Object tracking is an important part of a human-computer collaboration in a continuous

environment, in the sense of allowing the computer to obtain a better model of the real

world. For instance in the application area of autonomous vehicles where it is not possible

for a human to communicate the state of the environment accurately and quickly enough

given the requirements of the agent.

The broad area of application reflects the importance of reliable, exact, and effective object

tracking. There are several important steps towards effective object tracking, including the

choice of model to represent the object, and object tracking method suitable for the task.

Problem formulation:
Inability of people to analyse the image, process the digital image, and
recognize objects such as lines and simple geometric shapes like square,
rectangle, and circle.
The problem was about the ability to recognize different objects in a video,
where the objects can be rigid (e.g. container) as well as non-rigid (e.g.
clothing). Object tracking could
be used to automate the process of presenting information about a certain object
displayed.

Research question:
Why object tracking :
Creating a unique ID for each of the initial detections
Taking an initial set of object detections (such as an input set of bounding box
coordinates).
can be used to isolate features of a particular shape within an image.
Create virtual shape on the computer system.
relatively unaffected by image noise.

Objective:
 object tracking is the process of locating moving objects over time using
the camera in video sequences in real time. The objective of object
tracking is to associate target objects in consecutive video frames. Object
tracking requires location and shape or features of objects in the video
frames. Every tracking algorithm needs to detect moving object. So
object detection is the preceding step of object tracking in computer
vision applications. After that, detected object can be extracted by the
feature of moving object to track that moving object into video scene. It is
 challenging task in image processing to track the objects into consecutive
frames. Various challenges can arise due to complex object motion,
irregular shape of object, occlusion of object to object and object to scene
and real time processing requirements

 detect Geometric shapes.

 Creating a unique ID for each of the initial detections

Scope:
Since there exists many different approaches, methods, and variants of methods
for object tracking it would be an impossible task to thoroughly cover all
methods used in some object tracking implementation. Especially since there
are proprietary methods whose implementations and designs are not available
for study. This study focused on the, in literature, most common categories or
groups of objects tracking for rigid as well as non-rigid objects. Aside from
object tracking techniques this thesis does also include other important parts of
the tracking process. Since the aim of this thesis is to review methods for object
tracking.

Significance of study:
The importance of object tracking is reflected by the broad area of applications
such as video surveillance, human-computer interaction, and robot navigation.
The purpose of this study was to examine, evaluate, and make a sum- Mary of
the most common
 video surveillance:
System is basically used for analysis and explanation of object behaviours. It
consists of static and moving object detection, video tracking to understand the
events that occur in scene. The most important objective of this survey paper is
to determine the various methods in static and moving object detection as well
as tracking of moving objects. Any video scene contains objects that can be
determined by object detection technique. There are various classes of detected
object such as tree, clouds, person and other moving objects. Detection for
moving object is a very challenging for any video surveillance system. Object
Tracking is used to find the area where objects are available and shape of
objects in each frame in higher level application.

Motivation of the study:

 The motive of object tracking is to recognize and locate (localize) all
known objects in a scene.
 Since there exists many different approaches, methods, and variants of
methods for object tracking it would be an impossible task to thoroughly
cover all methods used in some object tracking implementation.
Especially since there are proprietary methods whose implementations
and designs are not available for study.
 associate target objects in consecutive video frames.

Literature Review:
associate target objects in consecutive video frame. Some of the first
occurrences of computer vision related work took place in the early 1970s.
Researchers wanted to be able to mimic human intelligence in computers and
computer
vision was considered a visual perception component in this agenda. The task of
solving
the vision input problem, i.e. feeding a computer with visual input and making it
describe
what it sees, was believed by some pioneers in the area of artificial intelligence
to be an easy
step along the way of solving more difficult and interesting problems [37]. One
example of
this is The summer vision project where the undergraduate student Gerald J.
Sussman was
asked by the Artificial Intelligence Group at MIT to perform a summer project.
The project
aimed to make a computer describe what it saw by linking a camera to it [25].
Turns out,
this task would have needed a couple of more decades of research to perform.
Digital image processing was already an existing field. However, the wish to get
a full
scene understanding brought forth the desire to use images to recover the three
dimensional
structure of the world, which was considered a step towards the ultimate goal.
Some of
the first attempts for this kind of scene understanding was done by edge
extraction and
translation of the 2D lines to 3D structures. During this time some work was
also performed
in interpreting factors such as color intensities and shade variations to be able to
explain
them in terms of image phenomena’s like surface orientation and shadows.
In the 1980s focus was shifted to performing quantitative image analysis. A lot
of effort was
put in developing sophisticated mathematical techniques for this purpose.
Research
was also performed to improve the ability to detect objects in images, with
methods like
edge and contour detection, as well as introducing the concept of evolving
contour
trackers such as snakes. Continued work made researchers come to the
discovery that
a lot of the proposed algorithms could be treated within the same optimization
framework.
In other words, they discovered that several algorithms could be described using
the same
mathematical framework, if it was posed as different optimization problems.
During the following decade researchers continued to explore relevant topics,
where some
became more interesting than others. A lot of important work, not least for the
ability to
track objects, was performed during this period when various tracking
algorithms were improved drastically. These included contour tracking using
active contours such as snakes,
and particle filters. Also, worth noticing is that interaction with computer
graphics increased. To explore the idea of creating animations using images of
the real world
the technique of image morphing was used, and later other techniques such as
view.
interpolation. A lot of progress and improvement was also done in important
fields such
as optical flow methods, global optimization using graph cut techniques, and
image segmentation. During the 2000s the work with developing techniques in
the area of computer
vision has continued. Among other areas we have seen increased interest in the
area of complex global optimization problems, where focus has been on
development of more efficient
algorithms.
Tracking objects in video has made great progress in some categories, for
instance humans,
faces, and animals. Despite this it remains a challenge to track generic objects
since their
visual appearance can change from one moment to another, due to, for example,
movement
or light changes. Even though we still have a long way to go and many areas to
explore further, there has been great progress of object tracking techniques
during the years.
Object tracking is today used in applications such as video surveillance, human-
computer
interaction, robot navigation, activity recognition, anomaly detection, virtual
reality, object
navigation, and path detection.

what is object tracking:

 the problem of estimating the trajectory of an object in an image plane as it
moves around a scene,
Object tracking is a discipline within computer vision, which aims to track
objects as they move across a series of video frames. Objects are often people,
but may also be animals, vehicles or other objects of interest, such as the ball in
a game of soccer. Below are impressive results achieved by SORT, a deep
learning object tracking algorithm.
Object tracking has many practical applications including surveillance, medical
imaging, traffic flow analysis, self-driving cars, people counting and audience
flow analysis, and human-computer interaction.
Technically, object tracking starts with object detection —identifying objects in
an image and assigning them bounding boxes. The object tracking algorithm
assigns an ID to each object identified in the image, and in subsequent frames
tries to carry across this ID and identify the new position of the same object.
There are two main types of object tracking:
Offline object tracking—object tracking on a recorded video where all the
frames, including future activity, are known in advance.
Online object tracking—object tracking done on a live video stream, for
example, a surveillance camera. This is more challenging because the algorithm
must work fast, and it is not possible to take future frames and combine them
into the analysis.
Object tracking vs object detection:
Object detection has evolved substantially in the past two decades, with the
move from traditional statistical or machine learning approaches to deep
learning approaches based on Convolutional Neural Networks (CNN). The
introduction of deep learning improved the accuracy and analysis power of
object detection by an order of magnitude.

To some, object tracking is simply an extension of object detection. The

creators of a popular algorithm called Simple Online and Realtime
Tracking (SORT) make the assertion that modern object detection algorithms
can do most of the work of detecting objects and re-identifying in subsequent
frames, and object tracking can be reduced to simple heuristics.

Others have developed extensive object training algorithms that work in tandem
with object detection, and apply deep learning techniques to carry over an
identified object into the next video frames.

Challenges of object tracking compared to static object detection:

 Re-identification—connecting an object in one frame to the same object

in the subsequent frames

 Appearance and disappearance—objects can move into or out of the

frame unpredictably and we need to connect them to objects previously
seen in the video

 Occlusion—objects are partially or completely occluded in some frames,

as other objects appear in front of them and cover them up

 Identity switches—when two objects cross each other, we need to discern

which one is which
  Motion blur—objects may look different due to their own motion or
camera motion

 Viewpoints—objects may look very different from different viewpoints,

and we have to consistently identify the same object from all perspectives

 Scale change—objects in a video can change scale dramatically, due to

camera zoom for example

 Illumination—lighting changes in a video can have a big effect on how

objects look and can make it harder to consistently detect them

Object Tracking Algorithm:

In this section, we’ll introduce three popular object tracking algorithms that use
deep learning methods: SORT, GOTURN and MDNet.
Simple online and real time tracking:

SORT is an object tracking algorithm that relies mainly on the analysis of an

underlying object detection engine. It can plug into any object detection
algorithm. The algorithm tracks multiple objects in real time, associating the
objects in each frame with those detected in previous frames using simple
heuristics. For example, SORT maximizes the IOU (intersection-over-union)
metric between bounding boxes in neighboring frames.
:Generic object tracking Using regression Network (goturn)
GOTURN is trained by comparing pairs of cropped frames from thousands of
video sequences. In the first frame (“previous frame”), the location of the object
is known, and the frame is cropped to twice the size of the bounding box around
the object, with the object centered.
The algorithm then tries to predict the location of the same object in the second
frame (“current frame”). The same double-sized bounding box is used to crop
the second frame. A Convolutional Neural Network (CNN) is trained to predict
the location of the bounding box in the second frame.

multi-domain Network (MDNet)

Multi Domain Network (MDNet) is a CNN architecture that won the VOT2015
challenge. The objective of MDNet is to speed up training in order to provide
real-time results. The strategy is to split the network into two parts. The first

part acts as a generic feature extractor that trains over multiple training sets and
learns to distinguish objects from their background. The second part is trained
on a specific training set and learns to identify objects within video frames.So
MDNet makes it possible to modify the weights of only the last few CNN layers
during training, reducing computation time significantly.
experimental result:

conclusion
The visual tracking algorithm for multiple object tracking based on Contourlet
transform works more efficiently than the standard blob tracking method which
is based on area and Centroid of the object. We introduced tracking method
based on the 3D color histogram for color feature extraction and tracking the
region. Region matching has been carried out using 2D seven invariant
moments calculated from the histogram, which needs to match only seven
descriptors of each region. So, the execution time taken by the algorithm is less
than the conventional matching methods. Also, to overcome the problem of
same color descriptor region, feature extraction using Contourlet transform has
been introduced effectively. Algorithm uses multiple methods for tracking the
object in efficient way, which can handle the color features as well as edge point
features.

References:
 https://missinglink.ai/guides/computer-vision/object-tracking-deep-learning/?
fbclid=IwAR0peFPgOP-
d_RGD2BKcCIb8psKDManMBT9sQBQ3FQzxtYS1wKvxG6cNlk4

 http://www8.cs.umu.se/education/examina/Rapporter/SannaAgrenFinal.pdf?
fbclid=IwAR0ZSoaDm4JqrdCv0EcziAltyxAKLMXS5Ay1x_2-kuRs9fRcKCZ2vVevGwc

 https://missinglink.ai/guides/computer-vision/object-tracking-deep-learning/?
fbclid=IwAR3YaTwr_JTGv8HF1LtnrlzDj2IjO913EGhnSxva-NCoX7sjQ1NO94aQPa4