vnese-id-extractor

Implemented by:

Ho Chi Minh City University of Technology - Control Engineering and Automation

1. Nguyen Ngoc Nhan
2. Thai Quang Nguyen
3. Pham Ngoc Tran
4. Nguyen Gia Khiem
5. Pham Binh Nguyen

Youtube video tutorial

Link: https://youtu.be/4DhUrMDltvE

How to run the code

Step 1: Install anaconda, python and git

Check out this link: https://youtu.be/ZRC2nzP1w_4

Step 2: Clone this repo

1. Open anaconda terminal, navigate to your directory
2. Run this command to clone the code:

git clone git@github.com:khiemnguyen240900/vnese-id-extractor.git

3. Then navigate to the code

cd vnese-id-extractor

Step 3: Create conda environment and install requirements

1. Create conda environment with cpu or gpu

# Change directory to installation
cd installation

# for CPU
conda env create -f conda-cpu.yml
conda activate yolov4-cpu

# for GPU
conda env create -f conda-gpu.yml
conda activate yolov4-gpu

2. Install requirements

# for CPU
pip install -r requirements.txt

# for GPU
pip install -r requirements-gpu.txt

3. Back to the base folder

cd ..

Step 4: Download and convert custom Yolo weights for id cards

1. Download my pre-trained weights at: https://drive.google.com/drive/folders/1TwrMzlOS2HuOv628ZOQeqANTQRpUapwh?usp=sharing
2. Put the weights file into: ./yolov4_card_detection/data/
3. Put the names file into: ./yolov4_card_detection/data/classes/
4. Open ./yolov4_card_detection/core/config.py, change line 15 to

__C.YOLO.BASE = os.getcwd().replace(os.sep, '/')

5. Convert the Yolo weights from darknet to tensorflow

cd yolov4_card_detection
python save_model.py --weights ./data/yolov4-cards.weights --output ./checkpoints/custom-416 --input_size 416 --model yolov4
cd .. 

6. Ensure the conversion is successful by checking ./yolov4_card_detection/checkpoints folder
7. Undo step 4.4 (change line 15 back to)

__C.YOLO.BASE = os.getcwd().replace(os.sep, '/') + "/yolov4_card_detection"

8. Note: to train with your custom data, check out this tutorial from theAIguys: https://youtu.be/mmj3nxGT2YQ

Step 5: Run the code

1. Please read Appendix A for information about flags
2. Run the following code (make sure you are in the base folder)

# using webcam with interactive mode
python main.py --weights /checkpoints/custom-416 --video 0 --interactive

# using phone camera through IP Webcam app and save the aligned image
python main.py --weights /checkpoints/custom-416 --camera_ip "YOUR-CAMERA-IP" --output

3. Extracted information and aligned image will be store in ./output folder

Appendix A: Most useful flags

alignment_process

--alignment_process: show alignment process (default: 'false')

camera ip

--camera_ip: camera ip for external camera

interactive

--interactive: interactive mode in card alignment (default: 'false')

output

--output: to save aligned image to output folder (default: 'false')

video

--video: path to input video or set to 0 for webcam (default: '0')

weights

--weights: path to weights file (default: '/checkpoints/yolov4-416')

Other flags

Other flags such as --iou, --model, --size,... can be read by running this command

python main.py --helpshort