Parking System License Plate Detection Based on Convolution Neural Networks GPU Optimization

EasyChair Preprint no. 7205

Abstract

This paper presents a parking system license plate detection based on convolution neural networks GPU optimization. When number of strides increased 2 by 2 it reduced memory allocation and reduced training time by half, however accuracy is also reduced from 80% to 75.79%. Accuracy is traded-off to avoid running into GPU memory allocation issues. Parking system license plate detection based on convolution neural networks (CNN) uses convolutional layers that are either completely interconnected or max pooled. The convolutional layer performs a convolutional operation on the input before passing the result to the next layer. The network can be much deeper due to this convolutional operation. With this, convolutional neural networks can be effective in image and video recognition, however it requires graphics processing unit GPU optimization to avoid running into memory issues.

Keyphrases: Convolution Neural Network, GPU optimization, Parking System License Plate Detection

@Booklet{EasyChair:7205,
  author = {Ziad Elkhatib and Adel Ben Mnaouer and Omar Mashaal and Nor Azman Ismail and Mohd Azman Bin Abas and Fuad Abdulgaleel},
  title = {Parking System License Plate Detection Based on Convolution Neural Networks GPU Optimization},
  howpublished = {EasyChair Preprint no. 7205},

  year = {EasyChair, 2021}}