License Plate Recognition System using YOLO and Mask R-CNN

Abstract

License plate devices have been commonly used in parking lots in recent years. The Traditional plate recognition  devices used on car park have fixed source of light and an angles for shooting in ordering to quickly distinguish license plates. Deformation oflicense recognition  plate can also be especially extreme for tilted angles like for examplelicense plate images taken with ultra-widened angle lens or it can be fisheye lens, resulting in poor identification of standard license plate recognition systems. Mask RCNN Device that couldalso be useful for different angles of shooting as well as oblique pictures. Experimental findings have shown that proposed architecture will be able to classify license plates which have bevel angles of over 0/60. Proposed Mask R-CNN system hadalso made substantial progress in character recognizingwhich are inclined more than 45 degrees in comparison with approach of using the YOLOv2 model. Results from experiments also show that the system proposed in the open data plates collection is superior to other methods (known as AOLP dataset).