Donget et al. [
18] proposed a classification of vehicle types with a Convolutional Neural Network (CNN) from the frontal view of the vehicles. The accuracy of their proposed approach reached 96. 1% during the day and 89.4% during the night. Kul et al. [
19] used the Support Vector Machine (SVM), Adaboost, and Artificial Neural Network (ANN) to classify vehicles. They obtained an accuracy of 87.5%, 81.6% and 85.4%, respectively. Kul et al. [
20,
21] proposed a two-stage vehicle classification system. Tashiev et al. [
22] developed a vehicle classification system that uses the YOLO real-time object classification framework and tested their work on the BITvehicle dataset. Their work reached an accuracy of 90.35%. They used TinyYOLO [
23] and two CNN; one for vehicle detection and the other for vehicle classification. Their method with Tiny-YOLO reached 89.19% accuracy in terms of the Intersection over Union (IOU) metric. Kul et al. [
24] showed that vehicle classification is a challenging problem due to the different dimensions of vehicles. Goncalves et al. [
25] have established an optical character recognition system with a convolutional neural network for plate recognition. Their proposed detection approach reached an accuracy of 79.3% , and their proposed recognition approach reached an accuracy of 85.6%. Laroca et al. [
26] identified the license plate recognition system based on the YOLO [
23] object detector. They trained the YoloV2 architecture, which is built on the Darknet [
27] library, with their own tagged data set. This model does not perform well when the frames have inclined angles. Their system performed with an accuracy rate of 78.33. Yonetsu et al. [
28] proposed a two-stage YOLOv2 for the detection of license plates. They extracted cars and license plates from images with objects of cars. They also built a database of Japanese license plates. Their proposed license plate detection method reached an accuracy of 87% in clear weather, 74% at nightfall, and 11% in the dark. In [
29], Nam et al. used the background modeling and subtraction (BGS) model for vehicle detection. To make their classifier robust, they performed the Gaussian functions of the OpenCV library. Z. Selmi and colleagues achieved an accuracy of 92.7%. Selmiet et al. [
30] proposed a deep learning license plate recognition system. Before recognizing the characters on a license plate, they built a CNN model for the classification of plates and nonplates. Their proposed method reached 94.8% accuracy with the Caltech dataset. Their system fails in some of the cases where there are multiple license plates. D. Pu et al. [
31,
32] presented a real-time CNN-based license plate detection system. But their approach did not work well for small license plates. Abdullah et al. [
33], presented a real-time license plate recognition system for Bangladesh. Their dataset includes photos that have different environmental conditions and is public. They used the YOLOv3 algorithm for their system and their method achieved 85% in terms of IoU for digit recognition. Silva et al. [
34] have developed a system that includes vehicle detection, license plate detection, and plate reading steps. Their approach allows reading license plates from different angles, and they also have created a synthetSilva ic data set.