Road Signs with AI powered applications
Session
Computer Science and Communication Engineering
Description
In this paper about Big Data we have experimented with the known neural network for image classification CNN (known as Convolutional Neural Network), where we build a model based on The German Traffic Sign Benchmark dataset and add certain configurations which include layers like convolutional, relu, pooling and fully connected layer. We continue training our model for a certain number of epochs, check the results and compare the performance by observing the values of accuracy and loss, during which time our model is improving itself through forward propagation and backpropagation, until we have a well-defined neural network that is good enough to detect features. The model accuracy we achieve is 96%. Since we achieved a good result on accuracy we continue on deploying our model on a desktop application with a simple graphical user interface which makes testing and using our model real easy and user friendly.
Keywords:
CNN, Traffic Sign Recognition, Neural Networks, Artificial Intelligence, Data Mining, Image Classification.
Proceedings Editor
Edmond Hajrizi
ISBN
978-9951-550-47-5
Location
UBT Kampus, Lipjan
Start Date
30-10-2021 12:00 AM
End Date
30-10-2021 12:00 AM
DOI
10.33107/ubt-ic.2021.389
Recommended Citation
Morina, Vesa and Hajrizi, Edmond, "Road Signs with AI powered applications" (2021). UBT International Conference. 406.
https://knowledgecenter.ubt-uni.net/conference/2021UBTIC/all-events/406
Road Signs with AI powered applications
UBT Kampus, Lipjan
In this paper about Big Data we have experimented with the known neural network for image classification CNN (known as Convolutional Neural Network), where we build a model based on The German Traffic Sign Benchmark dataset and add certain configurations which include layers like convolutional, relu, pooling and fully connected layer. We continue training our model for a certain number of epochs, check the results and compare the performance by observing the values of accuracy and loss, during which time our model is improving itself through forward propagation and backpropagation, until we have a well-defined neural network that is good enough to detect features. The model accuracy we achieve is 96%. Since we achieved a good result on accuracy we continue on deploying our model on a desktop application with a simple graphical user interface which makes testing and using our model real easy and user friendly.