Publication

Abstract : Recognition of cars is very important for the control and surveillance systems. Automobiles can be recognized by number plates, which contains a unique combination of alphabets and numbers. However, it's a hard and intensive job for humans to manually recognize all the parked or passing car number plates. In this paper, we approach a training-based pathway for vehicle number plate recognition. Most of the previous works in automatic number plate recognition (ANPR) systems have limitations in their working conditions, like for example restricting them to stationary backgrounds, indoor area, restricted vehicle speeds, prescribed driveways, fixed illumination, or match the predefined distance between camera and vehicle. The main objective of our work is to create a robust number plate recognition model that works under different illuminations and angles. We created our recognition model by training on our manually collected car number plate dataset using YOLO V3. The algorithm has been tested over 640 images which are of different colours, and illuminations.