ariesiitr/traffic-violation-detection

Inter IIT Tech Meet 12.0 Engineering Conclave

Problem Statement

To create a deep-learning-based system to detect and store information on traffic violators in a database hosted on a website to collect proof and number plates of the criminals simultaneously.

Mentors

Pranav Arya

Contributors

Riya Jindal

Ishika Arya

Rishika Pilania

Technology used

Approach

1) Learn different Computer Vision and Deep Learning models

2) As Yolo was chosen for the project, do an in-depth study of its working and custom training

3) Collect images using Google and Kaggle

4) Labeling of images was implemented using labelImg

5) Train the model on the collected Data set

6) Made a Django-based web interface to host the model

Solution

1) The initial dataset consisted of 100 images which were expanded using image processing techniques using OpenCV

2) We used labelImg to gentrate labled .txt file

3) Searching for proper training parameters was the toughest part, after multiple experiments and advice from mentors we were able to fix parameters like learning rate, no of steps for training, and much more

4) Model training was done using Google Collab and data was accessed via Google Drive

5) Initial problems in the model were resolved using better data sets, changing parameters

6) After training the model with sufficient accuracy it had to be hosted on a website

7) It was deployed on Django because it is written in Python, OpenCV can be used with Django and it has many inbuilt features to aid the process of integration and building the website

Working

Get API access

Go to https://platerecognizer.com/ and get your credentials

Put them in webcam\views.py and add credentials in line 23

Run the following commands

pip install -r requirements.txt

python manage.py create superuser (create a super user to access the admin page)   

python manage.py runserver

After starting the server

Go to localhost:8000/admin

in two-wheeler upload image to be tested

Go to localhost:8000/index/camera2 to run the test and see the object detection output

Go to localhost:8000/admin/webcam/crime2/ to view the crime result

Result

A platform was successfully presented to detect people on a two-wheeler not wearing helmets present in a given frame, storing proof of the violation and the number plate of the vehicle, leading to 30 points(max possible) in the Engineering conclave of Inter IIT Tech Meet 12 helping Institute to Bag Silver Medal for the same