license-plate-recognition

license plate recognition project using DNN

Task

• To recognize the Plate Number and Character

(1) First Step

• Train Data generation get_plate_data.ipynb
• generate the training image (Korean Version)
• Plate shows that string ( 99 가 1234 )
• Plate number generated on the random background images
• Thanks to Matt's blog Link, Github
• I made randomly generated training image (~15,000)
• Image size 512x512
• Label-coord ( ex, 99 가 1234 ) and mask for plate ( object detection task )

(2) Second Step

• Training the generated data to detect 'plate'

(3) Third Step (TBD)

• highlight the detect place
• then recognize the character (words and number)

Sample data images

• plate number : [ 97 지 0912 , 09 가 9871 ]

Model Training

model_train.py

Thanks to ( https://github.com/experiencor/keras-yolo2 )

• [ Devised date - 06/12/18 ]
• Detect the license plate
• Anchor Adjustment to find the various ratio of plate
• Successfully detect the plate in Test image (real world case )
• However still can't detect the small plate -> to be adjusted anchor size

Requirements

• Tensorflow-gpu (https://www.tensorflow.org/install/)
• Additional package

pip install -r requirements.txt