Smart Parking

Important Note:

Please note that this repository is forked from Priya Dwivedi's project on GitHub. You can find the her article here and her base source code here.

Contents:

1. webcam_test/ - Sample image to run and test the code on

2. train_data/ - Training data for CNN model. Sample data included

3. for_cnn/ - Directory to keep generated images for CNN. You have to manually label these images by separating them into sub-folders inside train_data/ as structure below. Please note that for_cnn/ is just only a directory to keep temporary images. It plays no role in main process.

.
├── for_cnn
└── train_data
    ├──train
    |   ├──empty
    |   └──occupied
    └── test
        ├──empty
        └──occupied

For the best training result, you should separate them with the ratio 8:2 for train:test

4. templates - Sample template to test API provided by Flask

5. camera.py - Python class for getting frames from your webcam

Change

self.video = cv2.VideoCapture(1)

to

self.video = cv2.VideoCapture(0)
or
self.video = cv2.VideoCapture('your_video.avi') # for static video

if you don't have an external webcam

6. CNN_model_for_occupancy.py - Python script for starting your model training process. In the first time you run the script, it may take a long time for downloading the model. Please note that I am using Priya Dwivedi's model and it works well on my data.

7. opencv_identifier.py - Python script for parking spot detection

8. main.py - Python script using Flask to provide API

9. config.json - Config file for dynamic data

10. read_data.py - Python script to read data from config.json. This script is imported in other Python files. You do not need to run this script.

11. webcam_utils.py - Python utils script that help you capturing video from webcams and save images for CNN model, etc. User's manual is written below.

12. image_utils - Python utils script that implements OpenCV framework's algorithms.

User's manual for webcam_utils.py

Run

python webcam_utils.py --mode

Where --mode:

1. capture - Start a video capturing instance. Press q to stop. Press c to toggle capturing mode. Video captured is save as outpy.avi

2. show-video - Show the video you captured as outpy.avi

3. save-a-frame - Read outpy.avi and save a random frame as webcam_test/test_frame.jpg

4. save-frames - Read outpy.avi and save many random frames into webcam_test/

5. get-training-data - Read all images in webcam_test/ and generate slot images into for_cnn/

Image processing steps

1. Original image

2. Selecting region of interested (ROI)

3. Perspective transformation

4. Image Resization

5. Image Bluration

6. Finding edges

7. Finding lines

8. Clustering blocks

9. Seperating parking slots

10. Predicting occupation

Installation

1. You would need Python 2.7 or 3.6 here

2. Please follow this instruction to install tensorflow or tensorflow-gpu, depends on your hardware.

3. Install opencv2

pip install opencv
or 
conda install opencv

4. Install flask version 0.12.2. Please note that higher flask versions could bring unexpected result.

pip install flask=0.12.2
or 
conda install flask=0.12.2

Run

1. Prepare training data with webcam_utils.py. Your prepared data will be in for_cnn/. However, please note that training data should be available in train_data/ with structure above. I have prepared some sample data for you in train_data/. Therefore, you may skip this step.

2. Run CNN_model_for_occupancy.py to start training model. Make sure you find file car1.h5 after training process.

3. Run main.py to start server. Your API should be available at http://0.0.0.0:8081/