An Efficient License Plate Detection Approach Using Lightweight Deep Convolutional Neural Networks

Table 1

Summary of features of the proposed method and recent methods for license plate detection.

Method	Purpose	Based on	Complexity	Accuracy	Advantages	Disadvantages
Tiny-YOLO [2]	Generic object detection	Deep learning	Very low	Low	Fast detection speed; easy to implement	Low accuracy
YOLOv2 [2]	Generic object detection	Deep learning	Medium	High	High accuracy; easy to implement	Struggles to detect small license plates
SSD-300 [3]	Generic object detection	Deep learning	High	High	High accuracy; easy to implement	High computational cost
R-FCN [4]	Generic object detection	Deep learning	High	High	High accuracy; easy to implement	High computational cost
Zhou et al. [5]	License plate detection	Traditional machine learning	High	High	High accuracy; no training process	Low detection speed; cannot detect license plates in difficult conditions
Li et al. [6]	License plate detection	Traditional computer vision	Very high	High	High accuracy; no training process	Very low detection speed; cannot detect license plates in difficult conditions
Yuan et al. [7]	License plate detection	Traditional computer vision	Very low	High	Very high detection speed; no training process	Cannot detect license plates in difficult conditions
Li et al. [8]	License plate detection and recognition	Deep learning	Medium	Very high	Very high accuracy; unified framework for license plate detection and recognition	Low detection speed
Nguyen et al. [9]	License plate detection	Deep learning	High	Very high	Very high accuracy	Low detection speed
Proposed method	License plate detection	Deep learning	Low	Very high	Very high accuracy; fast detection speed	Heavy data augmentation method is needed for training

Features are based on experimental results.