CMnet: A Compact Model for License Plate
Detection
Yao Cheng, Xin Zhou
(&)
, and Bo Ning
Dalian Maritime University, 1 Linghai Road, Dalian 116026, China
zhouxin314159@dlmu.edu.cn
Abstract. In recent years, with the rapid development of artificial intelligence
(AI), intelligent license plate (LP) detection which marks vehicle LP position
from the images has become one of the core technologies of intelligent transportation system. The traditional license plate detection algorithm has low
accuracy in some scenes, such as rotation, snow or fog, weak light, distortion,
and blur. Therefore, this paper proposes a model based on convolutional neural
network to realize vehicle LP detection. The model reduces the number of
convolution layers in the case of ensuring accuracy, thereby reducing the
number of parameters and saving time overhead. Through experiments, our
model achieved an accuracy of 98.8%, and the training time and detection time
were much less than the compare models.
Keywords: Object detection Á Convolutional neural network Á License plate
detection
1 Introduction
License plate detection [1–5] is an essential part of intelligent transportation systems
and is widely used in many traffic monitoring systems. The challenges of LP detection
have still existed in many current scenes. There are more or less problems in many
actually taken LP images, which makes it difficult to extract license plate features. Such
as Fig. 1a the license plate and the body of car are covered with snow in snowy days.
Figure 1b the shooting angle is distorted. Figure 1c blurring caused by distance. Figure 1d the lack of clarity caused by low light. Figure 1e uneven illumination caused by
strong light reflection. Many LP detections have lower accuracy in such complex
scenes, which need us to further explore the license plate detection under complex
scenes.
The license plate detection algorithm can be roughly divided into traditional
methods and neural network models. Traditional LP detection methods always use rich
edge information [6–12] or background color features [13, 14]. In [10], the authors
analyzed vertical edge gradients to select real license plate position. According to [14],
the author also used a color check module to help find the license plate position. The
breakthrough development of convolutional neural networks based on recent years has
stimulated the widespread application of object detection models in LP detection
problems [3, 4, 15]. Through computer efficient data processing and rich LP image
data, CNN improves the performance of LP detection by training the correlation
© Springer Nature Singapore Pte Ltd. 2020
Q. Liang et al. (Eds.): Artificial Intelligence in China, LNEE 572, pp. 576–583, 2020.
https://doi.org/10.1007/978-981-15-0187-6_68
Detection
Yao Cheng, Xin Zhou
(&)
, and Bo Ning
Dalian Maritime University, 1 Linghai Road, Dalian 116026, China
zhouxin314159@dlmu.edu.cn
Abstract. In recent years, with the rapid development of artificial intelligence
(AI), intelligent license plate (LP) detection which marks vehicle LP position
from the images has become one of the core technologies of intelligent transportation system. The traditional license plate detection algorithm has low
accuracy in some scenes, such as rotation, snow or fog, weak light, distortion,
and blur. Therefore, this paper proposes a model based on convolutional neural
network to realize vehicle LP detection. The model reduces the number of
convolution layers in the case of ensuring accuracy, thereby reducing the
number of parameters and saving time overhead. Through experiments, our
model achieved an accuracy of 98.8%, and the training time and detection time
were much less than the compare models.
Keywords: Object detection Á Convolutional neural network Á License plate
detection
1 Introduction
License plate detection [1–5] is an essential part of intelligent transportation systems
and is widely used in many traffic monitoring systems. The challenges of LP detection
have still existed in many current scenes. There are more or less problems in many
actually taken LP images, which makes it difficult to extract license plate features. Such
as Fig. 1a the license plate and the body of car are covered with snow in snowy days.
Figure 1b the shooting angle is distorted. Figure 1c blurring caused by distance. Figure 1d the lack of clarity caused by low light. Figure 1e uneven illumination caused by
strong light reflection. Many LP detections have lower accuracy in such complex
scenes, which need us to further explore the license plate detection under complex
scenes.
The license plate detection algorithm can be roughly divided into traditional
methods and neural network models. Traditional LP detection methods always use rich
edge information [6–12] or background color features [13, 14]. In [10], the authors
analyzed vertical edge gradients to select real license plate position. According to [14],
the author also used a color check module to help find the license plate position. The
breakthrough development of convolutional neural networks based on recent years has
stimulated the widespread application of object detection models in LP detection
problems [3, 4, 15]. Through computer efficient data processing and rich LP image
data, CNN improves the performance of LP detection by training the correlation
© Springer Nature Singapore Pte Ltd. 2020
Q. Liang et al. (Eds.): Artificial Intelligence in China, LNEE 572, pp. 576–583, 2020.
https://doi.org/10.1007/978-981-15-0187-6_68