REAL-TIME BUILDING INSTANCE DETECTION USING TENSORFLOW
BASED ON FACADE IMAGES FOR URBAN MANAGEMENT
A Thesis Presented to the Faculty of the
College of Information Technology and Computer Science
University of the Cordilleras
In Partial Fulfilment
of the Requirements for the Degree
BACHELOR OF SCIENCE IN DATA ANALYTICS
by
RODNEY ALCANTARA DOYAOEN
SAJIKO DULNUAN AROMIN
SOPHIA MARIE FERNANDEZ ARELLANO
March 2021
, APPROVAL SHEET
This dissertation entitled REAL-TIME BUILDING
INSTANCE DETECTION USING TENSORFLOW BASED ON FACADE IMAGES
FOR URBAN MANAGEMENT prepared and submitted by ARELLANO,
SOPHIA MARIE F., AROMIN SAJIKO D. AND DOYAOEN, RODNEY A. in
partial fulfilment of the requirements for the degree
BACHELOR OF SCIENCE IN DATA ANALYTICS, has been examined
and is recommended for acceptance and approval for oral
examination.
THELMA D. PALAOAG, DIT
Adviser
CAPSTONE PROJECT COMMITTEE
MELINDA A. BENINSIG, MIT THELMA D. PALAOAG, DIT
Member Member
JEFFREY S. INGOSAN, MCS
Chairperson
___________________________________________________________
PANEL OF EXAMINERS
APPROVED by the committee on Oral Examination on
__________________ with a grade of ______.
JEFFREY S. INGOSAN, MCS
Chairperson
MELINDA A. BENINSIG, MIT THELMA D. PALAOAG, DIT
Member Member
ACCEPTED AND APPROVED in partial fulfilment of the
requirements for the degree BACHELOR OF SCIENCE IN DATA
ANALYTICS.
THELMA D. PALAOAG, DIT JEFFREY S. INGOSAN, MCS
Program Chair Dean
College of Information College of Information
Technology and Computer Technology and Computer
Science Science
, ABSTRACT
1. Title: REAL-TIME BUILDING INSTANCE DETECTION USING
TENSORFLOW BASED ON FACADE IMAGES FOR URBAN
MANAGEMENT
1.1 Total No. of Pages : 158
1.2 Text No. of Pages : 76
2. Authors : Arellano, Sophia Marie Fernandez
Aromin, Sajiko Dulnuan
Doyaoen, Rodney Alcantara
3. Type of Document: Capstone Project
4. Type of Publication: Unpublished
5. Accrediting Institution: University of the Cordilleras
Gov. Pack Road, Baguio City
CHED-CAR
6. Keywords: Deep Learning, Object Detection, Computer
Vision, Tensorflow, Dissertation, Learning
7. Abstracts:
7.1 Rationale / Background of the Study
Development and improvement are the main contributors
to the success of a society. What regional authorities lack
are strategic planning and methods intended for urban
development and management. Without proper and
comprehensive plans, this brings the effect of ineffective
and unorganized allocation of resources and leads to
dissipated funds. To achieve urban management means
creating a better environment for the people and the
community. This study aims to develop a real-time building
instance detection that classifies different building types
based on facade images that shall aid urban planning and
management. The idea of this study, using object detection,
is defined as a computational technique that works with
identifying objects of a certain type in visual images and
videos. In line with this, a tool shall be used to perform
the task of real-time object detection. In order to test
the hypothesis if the model shall not successfully detect
and classify the various types of buildings within Baguio
, Abstract 4
City, Philippines, a pre-trained model SSD MobileNet
V2 FPNLite 320x320 with feature extractor from Tensorflow 2
Detection Model Zoo was configured to detect building
instances in real-time and iteratively optimized for better
accuracy. This study utilized the mean average precision
with two training methods, a simple split in the dataset
and a Stratified K-Fold cross-validation to measure the
accuracy of the model. Gaining an average of 0.47 or 47%
accuracy for the comparison of the simple split method. The
results suggest that even with limited data the stated
accuracy can be achieved. On this basis, other data can be
used to improve the detection performance.
The main aim of this study is to develop a real-time
building instance detection based on facade images using a
pre-trained model from Tensorflow 2 Detection Model Zoo.
7.2 Summary
The main aim of this study is to develop a real-time
building instance detection based on facade images. The
researchers sought to answer the following questions:
1. What are the processes needed to build the proposed
building type detection?
2. What model shall be adopted in the proposed
building type detection?
3. What is the degree of accuracy obtained using the
model?
4. What is the extent of usability of the proposed
building type detection?
This study adopted an agile software methodology which
is the Scrum Methodology. Through interviews and
researches, various user stories were gathered by the
researchers. Listing and validation of the possible
solutions to these problems were also done. After which,
the solutions were divided into smaller tasks arranging it
from most to least important. The tasks were then assigned
to specific individuals and daily meetings were held to
determine the current stage of the project each day. After
completing the tasks, the produced solutions were
demonstrated to potential users and stakeholders. Feedbacks
were taken into account and re-iterated previous processes
in order to come up with the desired result. Then, meetings
were held to identify the accomplishments done in each
phase, areas that needed to be improved and commitments to
achieve the next goal.
7.3 Findings of the Study
BASED ON FACADE IMAGES FOR URBAN MANAGEMENT
A Thesis Presented to the Faculty of the
College of Information Technology and Computer Science
University of the Cordilleras
In Partial Fulfilment
of the Requirements for the Degree
BACHELOR OF SCIENCE IN DATA ANALYTICS
by
RODNEY ALCANTARA DOYAOEN
SAJIKO DULNUAN AROMIN
SOPHIA MARIE FERNANDEZ ARELLANO
March 2021
, APPROVAL SHEET
This dissertation entitled REAL-TIME BUILDING
INSTANCE DETECTION USING TENSORFLOW BASED ON FACADE IMAGES
FOR URBAN MANAGEMENT prepared and submitted by ARELLANO,
SOPHIA MARIE F., AROMIN SAJIKO D. AND DOYAOEN, RODNEY A. in
partial fulfilment of the requirements for the degree
BACHELOR OF SCIENCE IN DATA ANALYTICS, has been examined
and is recommended for acceptance and approval for oral
examination.
THELMA D. PALAOAG, DIT
Adviser
CAPSTONE PROJECT COMMITTEE
MELINDA A. BENINSIG, MIT THELMA D. PALAOAG, DIT
Member Member
JEFFREY S. INGOSAN, MCS
Chairperson
___________________________________________________________
PANEL OF EXAMINERS
APPROVED by the committee on Oral Examination on
__________________ with a grade of ______.
JEFFREY S. INGOSAN, MCS
Chairperson
MELINDA A. BENINSIG, MIT THELMA D. PALAOAG, DIT
Member Member
ACCEPTED AND APPROVED in partial fulfilment of the
requirements for the degree BACHELOR OF SCIENCE IN DATA
ANALYTICS.
THELMA D. PALAOAG, DIT JEFFREY S. INGOSAN, MCS
Program Chair Dean
College of Information College of Information
Technology and Computer Technology and Computer
Science Science
, ABSTRACT
1. Title: REAL-TIME BUILDING INSTANCE DETECTION USING
TENSORFLOW BASED ON FACADE IMAGES FOR URBAN
MANAGEMENT
1.1 Total No. of Pages : 158
1.2 Text No. of Pages : 76
2. Authors : Arellano, Sophia Marie Fernandez
Aromin, Sajiko Dulnuan
Doyaoen, Rodney Alcantara
3. Type of Document: Capstone Project
4. Type of Publication: Unpublished
5. Accrediting Institution: University of the Cordilleras
Gov. Pack Road, Baguio City
CHED-CAR
6. Keywords: Deep Learning, Object Detection, Computer
Vision, Tensorflow, Dissertation, Learning
7. Abstracts:
7.1 Rationale / Background of the Study
Development and improvement are the main contributors
to the success of a society. What regional authorities lack
are strategic planning and methods intended for urban
development and management. Without proper and
comprehensive plans, this brings the effect of ineffective
and unorganized allocation of resources and leads to
dissipated funds. To achieve urban management means
creating a better environment for the people and the
community. This study aims to develop a real-time building
instance detection that classifies different building types
based on facade images that shall aid urban planning and
management. The idea of this study, using object detection,
is defined as a computational technique that works with
identifying objects of a certain type in visual images and
videos. In line with this, a tool shall be used to perform
the task of real-time object detection. In order to test
the hypothesis if the model shall not successfully detect
and classify the various types of buildings within Baguio
, Abstract 4
City, Philippines, a pre-trained model SSD MobileNet
V2 FPNLite 320x320 with feature extractor from Tensorflow 2
Detection Model Zoo was configured to detect building
instances in real-time and iteratively optimized for better
accuracy. This study utilized the mean average precision
with two training methods, a simple split in the dataset
and a Stratified K-Fold cross-validation to measure the
accuracy of the model. Gaining an average of 0.47 or 47%
accuracy for the comparison of the simple split method. The
results suggest that even with limited data the stated
accuracy can be achieved. On this basis, other data can be
used to improve the detection performance.
The main aim of this study is to develop a real-time
building instance detection based on facade images using a
pre-trained model from Tensorflow 2 Detection Model Zoo.
7.2 Summary
The main aim of this study is to develop a real-time
building instance detection based on facade images. The
researchers sought to answer the following questions:
1. What are the processes needed to build the proposed
building type detection?
2. What model shall be adopted in the proposed
building type detection?
3. What is the degree of accuracy obtained using the
model?
4. What is the extent of usability of the proposed
building type detection?
This study adopted an agile software methodology which
is the Scrum Methodology. Through interviews and
researches, various user stories were gathered by the
researchers. Listing and validation of the possible
solutions to these problems were also done. After which,
the solutions were divided into smaller tasks arranging it
from most to least important. The tasks were then assigned
to specific individuals and daily meetings were held to
determine the current stage of the project each day. After
completing the tasks, the produced solutions were
demonstrated to potential users and stakeholders. Feedbacks
were taken into account and re-iterated previous processes
in order to come up with the desired result. Then, meetings
were held to identify the accomplishments done in each
phase, areas that needed to be improved and commitments to
achieve the next goal.
7.3 Findings of the Study
