A New CNN-Based Method for Multi-Directional Car License Plate Detection

A New CNN-Based Method for Multi-Directional Car License Plate Detection


This paper presents a novel convolutional neural network (CNN) -based method for high-accuracy real-time car license plate detection. Many contemporary methods for car license plate detection are reasonably effective under the specific conditions or strong assumptions only. However, they exhibit poor performance when the assessed car license plate images have a degree of rotation, as a result of manual capture by traffic police or deviation of the camera. Therefore, we propose the a CNN-based MD-YOLO framework for multi-directional car license plate detection. Using accurate rotation angle prediction and a fast intersection-over-union evaluation strategy, our proposed method can elegantly manage rotational problems in realtime scenarios. A series of experiments have been carried out to establish that the proposed method outperforms over other existing state-of-the-art methods in terms of better accuracy and lower computational cost.

Existing System:

Thus, under the complex situation, it is relatively difficult to propose a robust method with hand-crafted features. Although people can employ multiple independent features and incorporate some models together, as mentioned, it is still hard to distinguish whether it is enough to meet the challenge with such limited features and models. To alleviate these problems, CNN-based methods have been devised, which automatically learn features from the acquired data.


These kinds of detection methods have recently yielded very impressive results however, their time consumption is significantly higher than those of the aforementioned techniques. Further, despite the viability of both traditional and CNN-based methods, the problems associated with multi-directional (MD) car license plate detection have not yet been satisfactorily resolved to date, because of the difficulties due to the viewpoint variation of hand-held cameras or the accidental rotation of mounted cameras.

Proposed System:

We propose a novel accurate rotation angle prediction method to realize multi-directional car license plate detection. To rapidly evaluate the intersection-over-union (IoU) between two rotational rectangles, we propose an approximate method, namely, the angle deviation penalty factor (ADPF).

To further promote the detection accuracy, we design a prepositive CNN model that is implemented before MD-YOLO, which serves to determine the “attention region” in the overall framework. The method of cascading the two models is based on prior knowledge: as the car license plates are fixed on the cars, some distance will inevitably exist between any two plates. The synergy of this concept is explained in the subsequent section.