A Novel Deep Learning Based Number Plate Detect Algorithm under Dark Lighting Conditions

Automatic number plate recognition (ANPR) is a meaningful and significant composition of intelligent traffic system, it is frequently used to record license plate automatically. Recently, most approaches to solve ANPR problems focused on improving the accuracy and efficiency in a typical way, they mainly consist localization, segmentation and recognition. However, these processes, especially the localization part, are deeply vulnerable to factors such as lighting conditions, orientation, rotation angle, etc. In this paper, distinct from many other methods, a model consisting computer visual processing and neural network recognition is introduced. It mainly contains color space detection and convolution neural network (CNN). After evaluation, this system is robust enough to process the detection under dark lighting conditions and achieve 98.3% accuracy based on 4500 number plate testing images, which is a relatively good result under dark conditions.