PROJECT - CS 662 SPRING 2023
Toward an automated approval system!
for transformed vehicles
Prepared for: Dr. Wadee Alhalabi
Prepared by: Maram Albsisi, 2301530
Atheer Aljohany, 2300042
22 February 2023
PROJECT - CS 662 SPRING 2023
Introduction
Transformed vehicle approval is burdensome for the Saudi Standards, Metrology, and
Quality Organization (SASO). Around ten thousand vehicles need approval from the authority per
year. This process requires shipping the vehicles to the center in Riyadh city. Or in other cases, the
authority sends an employee to a specific city to examine a group of vehicles and bear the costs.
Ultimately, these methods are expensive, complicated, time-consuming, and inefficient. For this
reason, investigating approaches to automatically or semi-automatically approve vehicles is critical.
Automated approval can be achieved using computer vision technology. In this project, we
propose a two-phased framework toward an automated approval system. The first phases focus on
data collection. Data collection is a very important step in the development of machine learning
models and computer vision classifications. This is because the accuracy of the model or
classification depends heavily on the data that has been collected. Quality data, with high coverage
of different aspects, allows the machine learning model or computer vision classification to
accurately identify objects or classify them into categories, leading to higher accuracy and better
results. In addition, data collection allows machines to learn patterns, which help them to make
predictions and decisions in the future. Thus, we put a great effort into the first phase. For the
second phase we propose a methodology using stat-of-art model for classification.
!
Proposed Solution
This section describes our methodology to approve transformed vehicles according to Saudi
standards automatically. two-phased framework. The first phase focuses on data collection and
labeling. Using a mobile application, we plan on collecting real-world images from customers
applying to the Saudi Standards, Metrology, and Quality Organization (SASO) for approval. At this
stage, data will be evaluated manually by experts. After collecting enough data, we move to phase
two. With the collected data, we train deep-learning model to classify approved vehicles. Finally, the
mobile application will be able to perform the full approval process automatically. Figure 1 shows the
two phased framework.
Figure 1. two phase framework toward automated approval system
The proposed methodology gather data from real users thus, we need interactive easy-to-use
application interfaces. Figure 2 shows the prototype interfaces. These interfaces were shown to the
authority for their feedback.
Figure 2. Application interfaces prototype for collecting data
The data collection application gathers details about the original vehicle ( type and size ), vehicle
details after transformation ( new type and new size ), and pictures of the vehicle ( side and front
view ). Table 1 shows an example of gathered data.
TABLE 1
ID org type org size new type new size front img side imp approved
1 truck medium car carrier medium id1_front.png id1_side.png yes
2 car carrier medium truck large id2_front.png id2_side.png no
… … … … … … … …
For the classification model, we suggest using a convolutional neural networks (CNN) model that
have achieved state-of-the-art performance in this task. CNN building block architecture is shown
( cf. figure 3 ).
The architecture contains; the Input layer which holds the pixel input, Convolutional layer it is the
operation that extracts data from an input layer and transfers to input into another input layers. The
pooling layer performs down sampling technique to an image input and reduces the size of the
image whilst preserving some of the important characteristics. The Rectified Linear Unit ( ReLU ) it is
the type of activation function used in the hidden layers of a DNN which can be formulated in
equation:
The Fully connected layer is the final process of CNN where basic ANN structure which occurs with
the use of the activation function to produce a numerical estimation value that represents the
prediction of the detected object.
Figure 3. Basic CNN architecture
Implementation Phase one
We create the application website using visual studio and XAMPP.
XAMPP is an abbreviation for cross-platform, Apache, MySQL, PHP and Perl, and it allows you to build
WordPress site offline, on a local web server.
Figure 4. used technology.
Figure 4. control panel of XAMPP.
code screenshots:
first page:
Figure 5. page 1 php source code.
Figure 6. page 1 css code.
Second page:
Figure 7. page 2 php source code.
Figure 8. page 2 css code.
Figure 9. code to added data for page 2 and link to page 3.
third page:
Figure 10. page 3 php source code.
Figure11. page 3 css code.
Figure 11. code to added data for page 3 and link to page 4.
Fourth page:
Figure 12. page 4 php source code.
Figure13. page 4 css code.
Figure 14. code to added data for page 4.
Figure 14. php code to show that data is loaded successfully.
The following code to identify a connection to the database:
Figure 15. php code to show that data is loaded successfully.
using phpMyAdmin we create our database and this is the attribute of table vehicle
Figure 16. created attribute.
Final product Screenshots:
Figure 17. page1.
Figure 18. page 2.
Figure 19. page 2 type of vehicle option.
Figure 20. page 2 size of vehicle option.
Figure 21. page 3.
Figure 22. page 4.
Figure 23. inform message.
Database:
Figure 24. database table .
lets say for example the company worker want to refuse the request for specific entry
,
they can edit the success attribute.
Figure 25. entry .
Figure 26. edit success value .
Figure [Link] edit.
website link on local host : [Link]
we the goal of this application is to collect data needed for first phase so it can be used in for future
development.