MishMash hackathon is India’s largest online diversity hackathon. The focus will be to give you, regardless of your background, gender, sexual orientation, ethnicity, age, skill sets and viewpoints, an opportunity to showcase your talent. The Hackathon is Live from 6:00 PM, 23rd March to 11:55 PM, 1st April, 2020
Before you start, please follow this format for your issue title:
TEAM NAME - PROJECT NAME - THEME NAME
ℹ️ Project information
You can select any one theme from - Mobility
Mobility, FinTech and Ed-Tech are Open Innovation Themes
XR has 3 sub-categories that you can choose from:- Epidemic, Urban, Remake
Deep Tech or Machine Learning is sponsored by UNILEVER and has 3 Problem Statements. If you pick this theme, you need to declare which problem statement you are going to work on
Social Impact has only 1 Problem Statement - So, if you pick this theme you just need to select this Theme and say you will work on the Problem Statement in the Idea Brief Field
Project Name: Give a suitable title to your project
PESU_ParkIt
Short Project Description:One line crisp description of your project
Detection of accuracy of parking done inside a predefined grid in parking lots and sending alert for the same.
Team Name: Please mention the same team name as mentioned over Skillenza
TECH_A_RATI
Team Members: Mention their Names & tag their GitHub handles
Deep Tech - Problem Statement - 3: If you have chosen to work on the problem statement - 3 then please submit both models based on the two datasets provided to you.
NA
Deep Tech - Problem Statement - 2: If you have chosen to work on the problem statement - 2 then please provide the reference for your dataset.
NA
Azure Services Used- Kindly mention the Azure Services used in your project.
Azure Notebook was used to perform training and testing of model. Azure Machine Learning was analysed for creation of Driver Rating System.
🔥 Your Pitch
Kindly write a pitch for your project. Please do not use more than 500 words
Everyone is extremely passionate about self driving cars and the future of mobility but one field of automation still goes unnoticed ,that is parking of these automobiles. The need for robust parking systems automated to such a degree that human intervention is minimal is increasing.
We present to you a system that does just that.
Our idea is to lead incoming cars to empty parking slots and use image processing to detect if the parking was done properly inside the grid predefined for that slot, if not an alert is sent to the guard's system.
The major steps in this project are:-
1) to detect the presence of a car
2) to detect grid of parking slot
3) calculate IoU
We detect the car using MaskRCNN , a pre-trained model named matterport was used for the same. We also do a comparison of MaskRCNN with YOLO that can detect objects faster but with comparatively lesser accuracy.
We then move on to detecting grids, there are various methods to do this, but are all based on camera positioning. Since we are using surveillance cameras we use pickeling to find the coordinates of a parking slot by creating a quadrilateral on a single video frame.
We then calculate IoU that is, Intersection over Union , this helps us detect if the grid is occupied by a car or not. Here we set the threshold in such a way, that if it is beyond a point we can say that the car is overlapping two grids and is therefore not parked properly. On this basis we rate the driver in range 1-5 and if the rating is less than 3 we alert the guard via SMS using twiliio API to go and help the driver park properly.
A further implementation of this project includes license plate detection of the cars that come into the parking lot and rate their parking . This rating can be used for subsequent parkings to decide what kind of parking slot should be given to this particular driver, such as corner spot, spot between 2 already parked cars, etc. based on the level of difficulty in parking and rating of driver.
🔦 Any other specific thing you want to highlight?
(Optional)
There can be a lot of noise in the data , such as frames containing cars entering or leaving a grid, we reject such frames in order to not create false alarms. We also use IoU because cars may overlap from the view of a camera that does not necessarily mean the car was parked wrong.
✅ Checklist
Before you post the issue:
[ *] You have followed the issue title format.
[* ] You have mentioned the correct labels.
[ *] You have provided all the information correctly.
Before you start, please follow this format for your issue title: TEAM NAME - PROJECT NAME - THEME NAME
ℹ️ Project information
PESU_ParkIt
Detection of accuracy of parking done inside a predefined grid in parking lots and sending alert for the same.
TECH_A_RATI
Ayushi Mohan https://github.com/Ayushi-Mohan Akarsh Shekhar https://github.com/akarshshekhar7
https://drive.google.com/drive/folders/19R07_3khcnZvjNHcq1b075RvlgAEmbC8?usp=sharing
https://github.com/akarshshekhar7/PESU-ParkIt
https://docs.google.com/presentation/d/1Ly-zPYYXUB5V8zKmjd7w1N-ED-TLA0p5Gq7wBBPnaXs/edit?usp=sharing
Azure Notebook was used to perform training and testing of model. Azure Machine Learning was analysed for creation of Driver Rating System.
🔥 Your Pitch
Kindly write a pitch for your project. Please do not use more than 500 words
Everyone is extremely passionate about self driving cars and the future of mobility but one field of automation still goes unnoticed ,that is parking of these automobiles. The need for robust parking systems automated to such a degree that human intervention is minimal is increasing.
We present to you a system that does just that. Our idea is to lead incoming cars to empty parking slots and use image processing to detect if the parking was done properly inside the grid predefined for that slot, if not an alert is sent to the guard's system.
The major steps in this project are:- 1) to detect the presence of a car 2) to detect grid of parking slot 3) calculate IoU
We detect the car using MaskRCNN , a pre-trained model named matterport was used for the same. We also do a comparison of MaskRCNN with YOLO that can detect objects faster but with comparatively lesser accuracy. We then move on to detecting grids, there are various methods to do this, but are all based on camera positioning. Since we are using surveillance cameras we use pickeling to find the coordinates of a parking slot by creating a quadrilateral on a single video frame.
We then calculate IoU that is, Intersection over Union , this helps us detect if the grid is occupied by a car or not. Here we set the threshold in such a way, that if it is beyond a point we can say that the car is overlapping two grids and is therefore not parked properly. On this basis we rate the driver in range 1-5 and if the rating is less than 3 we alert the guard via SMS using twiliio API to go and help the driver park properly.
A further implementation of this project includes license plate detection of the cars that come into the parking lot and rate their parking . This rating can be used for subsequent parkings to decide what kind of parking slot should be given to this particular driver, such as corner spot, spot between 2 already parked cars, etc. based on the level of difficulty in parking and rating of driver.
🔦 Any other specific thing you want to highlight?
(Optional)
There can be a lot of noise in the data , such as frames containing cars entering or leaving a grid, we reject such frames in order to not create false alarms. We also use IoU because cars may overlap from the view of a camera that does not necessarily mean the car was parked wrong.
✅ Checklist
Before you post the issue: