2 of 24

Drones have gained widespread interest in civilian and military applications, and can be even programmed to fly without the need of a human pilot

They can be equipped with multiple accessories, including cameras and other sensors to collect and transmit data in real time

Motivation :

4 of 24

Roos: Ever wonder how you can park on the campus grounds without a permit for the 20th time and not get a ticket? - I have!

University: Ever wonder how to issue a ticket to students without leaving your seat - we have!

5 of 24

The wait is now over…

6 of 24

EYE-CONIC

The only surveillance app you need

7 of 24

THE PROBLEM IN THE CURRENT SOLUTION

Due to the manual ticketing system, the hassle of walking around campus, checking for permits, and issuing a ticket has been an expensive and inaccurate process
This brings a lot of financial overhead to the University due to the number of cars that are parked on campus without a permit - additionally being an inconvenience to students with a parking permit

8 of 24

HOW DOES IT WORK?

Using drone technology, a mounted camera can detect the parked car’s license plate using object detection, process, and do real time ANPR (Automated Number Plate Recognition) and OCR (Optical Character Recognition)

The license plate number is then checked in the UMKC registered cars database - if the car is not registered, then the person is issued a ticket

9 of 24

$10.00/hr x 10 people for ~8 hours = $800

1 drone = $200 - $900 (one time payment)

10 of 24

3 Key Challenges :

Object detection, our object is a license plate.
Once object(license plate) is detected, extract the text from license plate.
Check if the parked vehicle is in the registered database.

1. Object detection

2. OCR

3. Registration Check

11 of 24

1. Object Detection

MobileNet model Object detection

Why MobileNet ?

It is tailored for resource constrained environment
Low memory requirement
High accuracy

12 of 24

Data set:

We took a dataset of license plates from all over the world.
So our model can be deployed anywhere.
We trained our model and achieved accuracy of 93%

500 images
80:20 Training-Test ratio
10,000 Steps

13 of 24

1. Object Detection (Continued..)

Problem :

The drone continuously read the frames and issued a ticket on every frame when OCR was performed. Thus, tickets were issued multiple times for the same person.

Solution :

We set a high confidence threshold for our object detection so only the correct plate is read. We implemented a counter which only takes in the first frame that passes our threshold. Finally, the license plate is then checked against the recent history.

14 of 24

Easy OCR architecture

EasyOCR uses a combination of automatic pattern recognition, feature extraction, and knowledge of text
EasyOCR can OCR text in 58 languages, including English, German, Hindi, Russian, and more

2. Easy OCR

15 of 24

3. Registration check

Once extracted text is received we check against our registered cars database and issue a ticket via email if plate is not registered.

16 of 24

Tech Stack

Deep Learning
Django
Angular
SQLite 3
DJI Tello SDK
Bootstrap

17 of 24

People Impact:

Our end users, such as organizations, having open public parking spaces with the problem of unregistered parking are benefited through a more confident surveillance system

We save operational expenses and scale at a more constant, less expensive rate, compared to manual labour

Our approach is more flexible when compared to traditional surveillance solutions in domains such as speed, accuracy and consistency

18 of 24

Ambition:

Our approach attacks a new problem: lackluster ticketing

It is difficult to consistently stay on top with the number of unregistered cars in large parking spaces for an average human

Many vehicles are missed and chances of human error are high

19 of 24

Innovation of Solution:

Our design breaks new ground in aerial surveillance by taking automated number plate recognition to the sky

Our solution also fully automates the existing manual ticketing system

20 of 24

Next phases

Other applications

Adding empty parking spot detection and updating it via a

mobile application to the registered users of the organization

to assist finding parking spots in large crowded areas

Agriculture

Security

21 of 24

Team that made it possible

22 of 24

Video demonstration of our project

https://www.youtube.com/watch?v=qrju8RLyib0

23 of 24

Thank you

Course : CS5590/490 - Python and Deep Learning

Coordinator : Lee, Yugyung

Instructor: Albishri, Ahmed

Course TA: Alqarni, Saeed

Organizers of Hack-A-Roo

1 of 24