NCTU

High accuracy and high reliability license plate recognition system

2017/09/13
CodeO17-CCL-10-1
Professorprofessor Cheng-Chang Lien
ApplicationCloud computing and the application of internet of things(IoTs), Surveillance and security, and Vehicle
FunctionImage detection and Image recognition
Technical BenefitDecrease error rate / improve stability, Easy operation/ reduce participation of operators, Improve efficiency/ increasing execution speed, Weather adaption, and Working on multi-camera at the same time
Technology StatusCan be transferred

The traditional license plate identification systems can have the following problems. First, the outdoor environments with complex backgrounds may reduce accuracy of the license plate locating and character recognition. Second, the license plate with large slant angles can make the license plate locating fail. In order to overcome these problems, we apply the novel deep learning technologies to generate a high robust license plate recognition system. By training the DNN models with a lot of databases constructed with different weather, lighting, and view angles, we can obtain a high accurate and stable license plate locating and recognition systems. First, we apply a deep learning DNN model to develop the license plate locating module and then apply another DNN model to identify the characters in the license plate. The proposed license plate recognition system can recognize the license plates with the accuracy 98% and the efficiency is about 15 fps in the outdoor environments.

(a) Low resolution image.

(b) Low contrast image.

(c) High tilt image.

demo video

Spec. request

platform :Window 7/8/10

tools: C/C++

hardware specifications: PC

demand:

CPU: Intel® Core™ i3 or higher

ram: at least 4GB

VGA: NVIDIA® GeForce® GTX 650  or higher