Deep Learning is awesome, it has given wings to computer vision and object detection/classification. We can even read number plates from images and automatically challan them if they are violating traffic rules. There are two ways of Doing number plate OCR. First way is to separate each character and then do OCR character by character. Other way is to feed the whole plate together. We need to see you you do in first approach. For test data, go to olx and download few sample car and bike images to test. You model/algorithm should be robust enough to work for one line and two line number plates, should be able to work on whilte, yellow and green background standard number plates.