Note: As we mentioned in today's mentoring session our doubts regarding number plates detecting less than 8 characters, the same was not commited to our web-app hosted on Heroku. The commit was done on this repository itself. but changes were not reflected on Heroku's git repo. Hence, the result fiasco on day2 evaluation.

Project is live check out here

• https://vehicle-detection.herokuapp.com
• https://vehicle-detection.herokuapp.com/streaming - for live streaming
• https://vehicle-detection.herokuapp.com/api - for Api

Vehicle Monitoring and Parking System

This is our implementation of parking and detecting based on YOLO approach done in Python 3. We have trained our model using YOLOv3, based on the Darknet framework for object detection (licence plate in this case) with the help of Google Colab. The training data can be found in the following directory: ./Dataset/

WE are cuurently Trying to improve our accuracy of the Ocr and testing models for the same

#####The repository includes:

• Yolo based approach to Detect Vehicle Number Plates.
• OCR and text extraction of detected Number Plate.
• Parking and Monitoring premised on recognized feature.
• Restricted zone parking alert.

####Results from training:

The columns in the given image represents Iterations, Loss of current iteration, average loss, learning rate, seconds to load, loaded images(in batch of 64) respectively. The mAP(mean Average Precision) achieved after 4100 iterations was 91.86% trained on over 420 images.

Requirements

• Python 3.x
  
• OpenCV 3.2.0 and other common packages listed in requirements.txt in child directories.

Getting Started

• Firstly we have prepared a Dataset which contains 420 images of cars which are taken at different angles.
• We have labelled our whole dataset for YOLO custom object detection.
• The input images are present in the Input images for training folder and the generated labels are present in YOLO labels.
• Training_Number_Plate.ipynb file is the training which is used to train YOLO for custom object detection like number plates.
• The Training file along with the dataset file is zipped into the folder and uploaded on the drive for training.
• We are using Google Colaboratory for training our model.
• Now that our model is trained we have to load our model for custom object detection.
• We have divided our project in two parts
  1.Vehicle Monitoring
  2.Vehicle Parking

1.Vehicle Monitoring

• Vehicle Monitoring means monitoring of vehicles and keeping a record of every vehicle that passes by.
• In our Monitoring system it will monitor whether a particular vehicle is a Residents Vehicle or Visitors Vehicle.
• Run the main.py file in Vehicle Monitoring directory.

# from main.py
net = cv2.dnn.readNetFromDarknet("./yolo trained weights/yolov3-obj.cfg",
                                 "./yolo trained weights/yolov3-tiny-obj_4000.weights")
net.setPreferableTarget(DNN_BACKEND_OPENCV)

• readNetFromDarknet() is used to load the trained model weights and configure file.
• setPreferableTarget() will instantiate this model on CPU.
• After the bounding box is generated we apply OCR(Optical Character Recognization) for the the rocognition of the characters.
• By applying OCR on number plate the value of Number plate is extracted and that value will be used to identify whether it is a Resident or a Vistor.
• For this we have created a fictional database of residents residing in a particular society.
• Now what happens is after the Number Plate is extracted, comparisons with database is made and processing is done accordingly.
• If the same Number Plate is found in the Database then it is a Resident otherwise it is a Visitor.
• After identifying as Visitor its record is maintained in our Visitor database with entry time.
• You can view our database entries by running resident.py in Database folder

####Results of Vehicle Monitoring(Identified as Resident):

####Results of Vehicle Monitoring(Identified as Visitor):

##2.Vehicle Parking

• We have divided the parking scenario into two parts
  i. Commercial Parking
  ii. Visitor Parking
  

####i.Commercial Parking

• In this scenario whenever you visit any public or commercial place you have to traverse whole parking in search of a parking spot.
• What our system does is that it shows which slot is available for parking and tells the person to park at that place and if the parking is full it will tell from the entry that parking is full so you dont have to traverse whole parking.
• When we run main.py file in Commercial Parking directory the YOLO model trained for custom object detection is loaded and it creates the bounding box around the number plate.
• After the bounding box is created the plate value is fetched by by applying OCR(Optical Character Recognition).
• After fetching the Number Plate it checks in Commercial Parking Database which slot is empty and it displays it.
• After Fetching the Number Plate if the parking is full and no slot is available it will show slot not available.
• You can view our database entries by running commercialparking.py in Database folder

####Results of Commercial Parking(Slots are available): 

####Results of Commercial Parking(Slots are not available):

####ii.Residential Parking

• Nowdays in Resident socities you are permanently alloted some parking slot and you park your vehicle at that place only and you pay for that place.
• What our system does in case of Residential Parking is that it classifies whether it is a resident or visitor.
• Run the main.py file in Residential Parking.
• Further steps are similar to the ones in aforementioned process of Commercial Parking.
• If he is a Resident then it displays all the details of the Resident along with the alloted slot value.
• If he is a Visitor a record is kept in Visitor database along with the fetched Number Plate and the Entry time of the vehicle.
• Currently only statically passed values seem to work. If some other person in the society has parked at your placed then by running Wrongslot.py you will come to know has parked at your alloted slot along with the contact number.

#wrongslot.py
slot_allot('GJ18BG5803',5)

• The contact Number of the wrong person will be displayed in the android app.
• You can view our database entries by running resident.py in Database folder
• If classified as visitor you can view our database entries by running visitor.py in Database folder

####Results of Residential Parking(Identified as Resident):

####Results of Residential Parking(Identified as Visitor):

####Results of Residential Parking(Someone has parked at your Slot):

####3.No Parking Solution

• It happens frequently that people deliberately park at restricted parking zones. So it is a necessity to prevent this from occuring chronically. We have put forward a solution that tries to address this scenario.
• Run the main.py in the No Parking Solution directory to get started.
• A window will pop up with an image of the parking lot, create bounding boxes. These boxes will indicate the restricted parking zones.
• You can change the video/image by editing the arguments passed in the main file as given below

# --image images/parkinglotimage.png  --data data/coordinates_1.yml --video videos/your_video_file.mp4 --start-frame 1

• (PROPOSED) As soon as these restricted parks are acquired, an alert can be sent to nearby towing authority/RTO officer or penalty can be charged on that vehicle by RTO officals. Note: For this to work, number plate extraction is to be done.

####Landing Page of our webapp:

####Signup Functionality in our Webapp:

####Functionality to accurately Detect license plates in our webapp:

####Commercial Parking Scenario in our webapp:

####Results of Residential Parking in our webapp and also visitors log:

####Results of our Api which we will expose to our end users:

####Results of our Streaming feature in our webapp:

####Added the chat feature to offer direct technical support to users:

####Can Detect car as well as Only Number plate:

####Added the chat feature for convinience of the users and added multilanguage support for every region of India:

References

• http://www.inf.ufpr.br/rblsantos/papers/laroca2018robust.pdf
• https://openaccess.thecvf.com/content_ECCV_2018/papers/Sergio_Silva_License_Plate_Detection_ECCV_2018_paper.pdf