libalpr locates and recognizes license plates in images. A convnet is used to generate a feature map of the image, and an LSTM reads the license plate character-by-character by paying attention to the part of the feature map corresponding to each character. The network architecture is fully convolutional, so it can accept images of arbitrary size, but it looks for number plates in windows of size 94x54 px.
Try libalpr with the pre-trained model: python demo.py
To train the model, a synthetic dataset of number plates is generated,
but font files and background images are required (location specified
in train.py).
- Python 2.7
- PyTorch 0.3
- Pillow
- Numpy