IMAGE DETECTION FROM VIDEOS USING
AI construct the combined features.
Abstract - How to accurately and effectively identify
people has always been an interesting topic, both in
research and in industry. With the rapid development
of artificial intelligence in recent years, facial
recognition gains more and more attention.
Compared with the traditional card recognition,
fingerprint recognition and iris recognition, face
recognition has many advantages, including but limit
to non-contact, high concurrency, and user friendly.
It has high potential to be used in government, public
facilities, security, e-commerce, retailing, education
and many other fields. Deep learning is one of the
new and important branches in machine learning.
Deep learning refers to a set of algorithms that solve
various problems such as images and texts by using
various machine learning algorithms in multi-layer
neural networks. Deep learning can be classified as a
neural network from the general category, but there
are many changes in the concrete realization. At the
core of deep learning is feature learning, which is
designed to obtain hierarchical information through
hierarchical networks, so as to solve the important
problems that previously required artificial design
features. Deep Learning is a framework that contains
several important algorithms. For different
applications (images, voice, text), you need to use
different network models to achieve better results.
With the development of deep learning and the
introduction of deep convolutional neural networks,
the accuracy and speed of face recognition have
made great strides. However, as we said above, the
results from different networks and models are very
different. In this paper, facial features are extracted
by merging and comparing multiple models, and then
a deep neural network is constructed to train and
CHAPTER 1: Introduction
Ever since IBM introduced first personal computer
on 1981, to the .com era in the early 2000s, to the
online shopping trend in last 10 years, and the
Internet of Things today, computers and information
technologies are rapidly integrating into everyday
human life. As the digital world and real world merge
more and more together, how to accurately and
effectively identify users and improve information
security has become an important research topic. Not
only in the civil area, in particular, since the 9-11
terrorist attacks, governments all over the world have
made urgent demands on this issue, prompting the
development of emerging identification methods.
Traditional identity recognition technology mainly
rely on the individual’s own memory (password,
username, etc.) or foreign objects (ID card, key, etc.).
However, whether by virtue of foreign objects or
their own memory, there are serious security risks. It
is not only difficult to regain the original identity
material, but also the identity information is easily
acquired by others if the identification items that
prove their identity are stolen or forgotten. As a
result, if the identity is impersonated by others, then
there will be serious consequences. Different from
the traditional identity recognition technology,
biometrics is the use of the inherent characteristics of
the body for identification, such as fingerprints,
irises, face and so on. 6 Compared with the
traditional identity recognition technology, biological
features have many advantages, as: 1.
Reproducibility, biological characteristics are born
with, cannot be changed, so it is impossible to copy
other people's biological characteristics. 2.
Availability, biological features as part of the human
body, readily available, and will never be forgotten.
3. Easy to use. Many biological characteristics will
not require individuals to corporate with the examine
device. Based on the above advantages, biometrics
has attracted the attention of major corporations and
research institutes and has successfully replaced
traditional recognition technologies in many fields.
And with the rapid development of computer and
artificial intelligence, biometrics technology is easy
to cooperate with computers and networks to realize
automation management, and is rapidly integrating
into people's daily life. When comparing the
differences between different biometrics, we can see
that the cost of facial recognition is low, the
acceptance from user is easy, and the acquisition of
information is easy. Facial recognition is the use of
computer vision technology and related algorithms,
from the pictures or videos to find faces, and then
analysis of the identity. In addition, further analysis
of the acquired face, may conduct some additional
attributes of the individual, such as gender, age,
emotion, and etc.
APPLICATION OF THIS PROJECT
Object detection is a computer technology related
to computer vision and image processing that deals
with detecting instances of semantic objects of a
certain class (such as humans, buildings, or cars) in
digital images and videos. Object detection has
applications in many areas of computer vision,
including image retrieval and video surveillance.
IDENTIFICATION OF CLASS & NEED
It is widely used in computer vision tasks such as
image annotation, vehicle counting, activity
recognition, face detection, face recognition, video
object co-segmentation. It is also used in tracking
objects, for example tracking a ball during a football
match, tracking movement of a cricket bat, or
tracking a person in a video.
Every object class has its own special features that
helps in classifying the class for example all circles
are round. Object class detection uses these special
features. For example, when looking for circles,
objects that are at a particular distance from a point
(i.e. the center) are sought. Similarly, when looking
for squares, objects that are perpendicular at corners
and have equal side lengths are needed. A similar
approach is used for face identification where eyes,
nose, and lips can be found and features like skin
color and distance between eyes can be found.
RELEVANT CONTEMPORARY ISSUES
The main disadvantages of manual attendance
tracking are the risk of human error. We're all human
and we can easily make mistakes when collecting
data manually. Manually tracking is also time
consuming and also increases the paper load.
PROBLEM IDENTIFICATION
With a face recognition attendance system, the entire
environment is automated. You won't just take the
attendance but also automatically record the entry-
exit time of the employees. It also adds to the security
of the workplace as the system can recognize who
left the designated area and when accurately.
TASK IDENTIFICATION
This project deals on face recognition where each
student in the class will be video graphed and their
attendance will be stored in a server. The teacher can
then record the attendance by just clicking some
pictures of the classroom. The system will recognize
the faces and mark their attendance.

CHAPTER 2: Literature survey
2.1 FACE DETECTION AND FACE TRACKING
This article Robust Real-time Object Detection [2] is
the most frequently cited article in a series of articles
by Viola that makes face detection truly workable.
We can learn about several face detection methods
and algorithm from this publication. The article Fast
rotation invariant multi-view face detection based on
real Adaboost [3] for the first time real adaboost
applied to object detection, and proposed a more
mature and practical multi-face detection framework,
the nest structure mentioned on the cascade structure
improvements also have good results. The article
Tracking in Low Frame Rate Video: A Cascade
Particle Filter with Discriminative Observers of
Different Life Spans [4] is a good combination of
face detection model and tracking, offline model and
online model, and obtained the CVPR 2007 Best
Student Paper. 10 The above 3 papers discussed
about the face detection and face tracking problems.
According to the research result in these papers, we
can make real time face detection systems. The main
purpose is to find the position and size of each face in
the image or video, but for tracking, it is also
necessary to determine the correspondence between
different faces in the frame.
2.1.2 FACE POSITIONING AND ALIGNMENT
Earlier localization of facial feature points focused on
two or three key points, such as locating the center of
the eyeball and the center of the mouth, but later
introduced more points and added mutual restraint to
improve the accuracy and stability of positioning
Sex. The article Active Shape Models-Their Training
and Application [5] is a model of dozens of facial
feature points and texture and positional relationship
constraints considered together for calculation.
Although ASM has more articles to improve, it is
worth mentioning that the AMM model, but also
another important idea is to improve the original
article based on the edge of the texture model. The
regression-based approach presented in the paper
Boosted Regression Active Shape Models [6] is
better than the one based on the categorical apparent
model. The article Face Alignment by Explicit Shape
Regression [7] is another aspect of ASM
improvement and an improvement on the shape
model itself. Is based on the linear combination of
training samples to constrain the shape, the effect of
alignment is currently seen the best. The purpose of
the facial feature point positioning is to further
determine facial feature points (eyes, mouth center
points, eyes, mouth contour points, organ contour
points, etc.) on the basis of the face area detected by
the face detection / tracking, s position. These 3
articles show 11 the methods for face positioning and
face alignment. The basic idea of locating the face
feature points is to combine the texture features of the
face locals and the position constraints of the organ
feature points.
2.1.3 FACE FEATURE EXTRACTION
PCA-based eigenfaces [8] are one of the most classic
algorithms for face recognition. Although today's
PCA is more used in dimensionality reduction in real
systems than classification, such a classic approach
deserves our attention. The article Local Gabor
Binary Pattern Histogram Sequence (LGBPHS): A
Novel Non-Statistical Model for Face Representation
and Recognition [9] is close to many mature
commercial systems. In many practical systems, a
framework for extracting authentication information
is PCA and LDA. Using PDA to reduce matrix to
avoid the matrix singularity problem of LDA solving,
then using LDA to extract the features suitable for
classification, To further identify the various original
features extracted after the decision-level fusion.
Although some of the LFW test protocols are not
reasonable, there is indeed a face recognition library
that is closest to the actual data. In this article,
Blessing Dimensionality: Highdimensional Feature
and Its Efficient Compression for Face Verification
[10], the use of precise positioning point as a
reference to face multi-scale, multi-regional
representation of the idea is worth learning, can be
combined with a variety of representation. The above
3 papers discussed about facial feature
positioning/alignment. Facial feature extraction is a
face image into a string of fixed-length numerical
process. This string of numbers is called the "Face
Feature" and has the ability to characterize this face.
Human face to mention 12 the characteristics of the
process of input is "a face map" and "facial features
key points coordinates", the output is the
corresponding face of a numerical string (feature).
Face to face feature algorithm will be based on facial
features of the key point coordinates of the human
face pre-determined mode, and then calculate the
features. In recent years, the deep learning method
basically ruled the face lift feature algorithm, In the
articles mentioned above, they showed the progress
of research in this area. These algorithms are fixed
time length algorithm. Earlier face feature models are
larger, slow, only used in the background service.
However, some recent studies can optimize the
model size and operation speed to be available to the
mobile terminal under the premise of the basic
guarantee algorithm effect.
CHAPTER 3: Design flow / Process
CONCEPT GENERATION
Every object class has its own special features that
helps in classifying the class for example all circles
are round. Object class detection uses these special
features. For example, when looking for circles,
objects that are at a particular distance from a point
(i.e. the center) are sought. Similarly, when looking
for squares, objects that are perpendicular at corners
and have equal side lengths are needed. A similar
approach is used for face identification where eyes,
nose, and lips can be found and features like skin
color and distance between eyes can be found.
SPECIFICATIONS / FEATURES
Face recognition is among the most productive image
processing applications and has a pivotal role in the
technical field. Recognition of the human face is an
active issue for authentication purposes specifically
in the context of attendance of students. Attendance
system using face recognition is a procedure of
recognizing students by using face biostatistics based
on the high definition monitoring and other computer
technologies.
The proposed system makes the use of opencv-
python, face_recognition, numpy, cmake and dlib
python-based packages to detect face biostatistics.
After face recognition attendance reports will be
generated and stored in excel format.
IMPLEMENTATION PLAN
Steps to Build the Face Recognition System
1. INSTALL LIBRARIES
 We need to install three libraries in order to
implement face recognition and the libraries
are as follows:
a) opencv-python: A library for some image
pre-processing

b) dlib: It is a modern C++ toolkit containing

machine learning algorithms
and tools for creating complex
software in C++ to solve real-world
problems.

As you can see RGB looks natural so you

will always change the channel to RGB.
c) face_recognition: The face_recognition
library, created and maintained by Adam 3. FIND FACE LOCATION AND DRAW
Geitgey that wraps around dlib facial BOUNDING BOXES
recognition functionality. You need to draw a bounding box around
the faces in order to show if the human face
has been detected or not.

2. LOADING IMAGES
After importing libraries you need to load an
image.
face_recognition library loads images in the
form of BGR, in order to
print the image you should convert it into
RGB using OpenCV.
 face_recognition.compare_faces
returns True if the person in both
images are the same other it returns
False.
4. TRAIN AN IMAGE FOR FACE
RECOGNITION Steps to Build the Face Recognition Attendance
This process is done in two steps and the System
steps are as follows:
a) Training 1. Install necessary libraries
At this stage, we convert the train
image into some encodings and
store the encodings with the given
name of the person for that image.

b) Testing
For testing, we load an image and
2. Define a folder path where your training
convert it into encodings, and now
image dataset will be stored
match encodings with the stored
encodings during training, this
matching is based on finding
maximum similarity. When you
c) Create a list to store person_name
find the encoding matching to the
and image array.
test image you get the name
d) Traverse all image file present in
associated with train encodings.
path directory, read images, and
append the image array to the
image list, and file-name to
classNames.
3. Create a function to encode all the train
images and store them in a variable
encoded_face_train.
4. Creating a function that will create a
5.
Attendance.csv file to store the attendance
with time, you need to create Attendance.csv
file manually and give the path in the
function
with open(“Attendance.csv”,’r+’) creates a
file and ‘r+’ mode is used to open a file for
reading and writing.
We first check if the name of the attendee is
already available in attendance.csv we won’t
write attendance again.
If the attendee’s name is not available in
attendance.csv we will write the attendee
name with a time of function call.
Read Webcam for Real-Time Recognition
a) Resize the image by 1/4 only
for the recognition part. output
frame will be of the original
size.
b) Resizing improves the Frame
per Second.
c) face_recognition.face_location
s() is called on the resized
image(imgS) for face bounding
box coordinates must be
multiplied by 4 in order to
overlay on the output frame.
d) face_recognition.distance()
returns an array of the distance
of the test image with all
 images present in our train constraint. There are several challenges that
directory. are faced by the Facial Recognitions System
e) The index of the minimum are as follows:
face distance will be the
matching face. 1. Illumination: It changes the face
f) After finding the matching appearance drastically, it is
name we call the observed that the slight changes in
markAttendance function. lighting conditions cause a
g) Draw bounding box using significant impact on its results.
cv2.rectangle().
h) We put the matching name on 2. Pose: Facial Recognition systems
the output frame using are highly sensitive to the pose,
cv2.putText(). which may result in faulty
recognition or no recognition if the
database is only trained on frontal
face view.

3. Facial Expressions: Different

expressions of the same individual
are another significant factor that
needs to be taken into account.
Modern Recognizers can easily
deal with it though.

4. Low Resolution: Training of

recognizer must be done on a good
resolution picture, otherwise the
model will fail to extract features.

5. Aging: With increasing age, the

CHAPTER 4: Results analysis and human face features shape, lines,
texture changes which are yet
validation
another
Although building facial recognition seems
easy it is not as easy in the real world
images that are being taken without any
 RESULTS
The program was able to detect the faces
and some objects in the video and also it
was able to name them correctly.

OUTPUTS
The Screenshots of the outputs are given
below:

CHAPTER 5: Conclusion and future work

CONCLUSION
In this project, we discussed how to create a face
recognition system using the face_recognition library
and made an attendance system.

The expected result of the proposed program was to

capture the faces of the students from the webcam
and record their attendance. But, unfortunately the
“dlib library” is very unstable and a problematic
library and due to these reason the program
sometimes does not even start. There is less chances
that the program will work. The reason could be as
we don’t have a high performance GPU or due to the
operating system or due to the python interpreter we
used.
FUTURE WORK
When building a facial recognition model, there are
many parameters which can be tuned to increase the
model performance. We can keep tuning our models
to increase its accuracy. Also, we will find a good
alternative of “dlib” library and implement in our
project as this library is somewhat problematic to use.
For a trained base model, we can re-train it using a
specific dataset. So another way to increase the whole
system’s performance is to capture the specific
people’s images and re-train the model based on this
small dataset. For example, if an organization with
3000 people uses this system, the model can be
trained to be very accurate on these 3000 people.
ACKNOWLEDGEMENT
We would like to express our sincere gratitude to
the professor because washing head of
Department Academic Unit 1 for providing us
with this wonderful opportunity of designing this
marvelous Independent project on “IMAGE
DETECTION FROM VIDEOS USING AI”
We are highly indebted to Mr. Rohit Kumar, our
mentor for his preserving guidance and
unflinching support in the process of crafting this
project and documenting this report, which is
worthy experience.
We would convey our special thanks to Mr Bilal
Bashir, for his substantial support and valuable
feedback, which paved the way for improvisations
and changes, helping uplifting the quality of their
design project.
Last but definitely not the least, we thank our
classmates of PH21BCS-404(A) for their constant
motivation and competitive inspiration
REFER
ENCES
1. [1]"cookbook.fortinet.com," 10 10 2018.
[Online]. Available:
https://cookbook.fortinet.com/face-
recognition-configuration-in-forticentral/.
[Accessed 10 10 2018].
2. [2]M. J. Paul Viola, "Robust Real-time
Object Detection," International Journal of
Computer Vision, pp. 137-154, 2004.
3. [3]H. A. C. H. a. S. L. Bo Wu, "Fast rotation
invariant multi-view face detection based on
real Adaboost," Sixth IEEE International
Conference on Automatic Face and Gesture
Recognition, pp. 79-84, 2004.
4. [4]H. A. Y. T. S. L. Yuan Li, "Tracking in
Low Frame Rate Video: A Cascade Particle
Filter with Discriminative Observers of
Different Life Spans," IEEE Transactions on
Pattern Analysis and Machine Intelligence,
pp. 1728-1740, 2008.
 Conference on Computer Vision and Pattern
Recognition, Portland, OR, 2013.

5. [5] C. J. T. D. H. C. A. J. G. T. F. COOTES,
"Active Shape Models-Their Training and
Application," COMPUTER VISION AND 11. [11] "Convolutional_neural_network,"
IMAGE UNDERSTANDING, pp. 38-59, 2017. [Online]. Available:
1995. https://en.wikipedia.org/wiki/Convolutional
_neural_network.

6. [6] D. C. a. T. Cootes, Boosted Regression 12. [12] S. S. Liew, "Research Gate," 1 3 2016.
Active Shape Models, BMVC, 2007. [Online]. Available:
https://www.researchgate.net/figure/Archite
cture-of-the-classical-LeNet-5-

7. [7] Y. W. F. W. a. J. S. X. Cao, "Face CNN_fig2_299593011. [Accessed 10 10

alignment by Explicit Shape Regression," in 2018].

2012 IEEE Conference on Computer Vision

and Pattern Recognition, Providence, RI, 13. [13] L. B. Y. B. a. P. H. Y. LeCun,
2012. "Gradient-based learning applied to
document recognition," Proceedings of the

8. [8] M. T. a. A. Pentland, "Eigenfaces for IEEE, pp. 1-45, 11 1998.

recognition," Journal of Cognitive 14. [14] xlvector, "JIANSHU," 25 7 2016.

Neuroscience, pp. 71-86, 1991. 47 [Online]. Available:
https://www.jianshu.com/p/70a66c8f73d3.
[Accessed 18 9 2018].

9. [9] S. S. W. G. X. C. a. H. Z. Wenchao
Zhang, "Local Gabor binary pattern 15. [15] F. S. Samaria, Face recognition using
histogram sequence (LGBPHS): a novel Hidden Markov Models, Doctoral thesis,
non-statistical model for face representation 1995.
and recognition," Tenth IEEE International
Conference on Computer Vision, pp. 786-
791, 2005.
16. [16] E. H. G. R. A. L. H. Learned-Miller, "
Labeled Faces in the Wild: A Survey,"
10. [10] X. C. F. W. a. J. S. D. Chen, "Blessing Advances in Face Detection and Facial
of Dimensionality: High-Dimensional Image Analysis, pp. 189-248, 2016.
Feature and Its Efficient Compression for
Face Verification," in 2013 IEEE
17. [17] V. Chu, "Medium.com," 20 4 2017. Available:

[Online]. Available: http://ispl.korea.ac.kr/~wjhwang/project/201

https://medium.com/initializedcapital/bench 0/TIP.html.

marking-tensorflow-performance-and-cost-
across-different-gpu-options69bd85fe5d58.
[Accessed 19 9 2018]. 48

18. [18] B. L. M. S. B. Amos, "Openface: A

general-purpose face recognition library
with mobile applications," CMU School of
Computer Science, Tech. Rep., 2016.

19. [19] B. Hill, "HOTHARDWARE," 20 8

2018. [Online]. Available:
https://hothardware.com/news/nvidia-
geforce-rtx-1080-rtx-1080-ti-799-1199-
september20th. [Accessed 10 9 2018].

20. [20] 4psa, "4psa.com," 28 6 2013. [Online].

Available: https://blog.4psa.com/the-
callbacksyndrome-in-node-js/. [Accessed 11
8 2018].

21. [21] W.-S. Chu, "Component-Based

Constraint Mutual Subspace Method," 2017.
[Online]. Available:
http://www.contrib.andrew.cmu.edu/~wschu
/project_fr.html.

22. [22] W. Hwang, "Face Recognition System

Using Multiple Face Model of Hybrid
Fourier Feature under Uncontrolled
Illumination Variation," 2017. [Online].