Deep Learning-Based Vehicle Classification for Low Quality Images.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Vehicle classification is crucial for intelligent transportation systems.
• Standard security cameras often capture low-resolution, distant traffic images.
• Existing models may be too complex for real-time processing of low-quality data.

Purpose of the Study:

• To propose a simple convolutional neural network (CNN)-based model for vehicle classification.
• To evaluate the model's effectiveness on tiny, low-resolution vehicle images.
• To compare the proposed model with established VGG16-based CNNs.

Main Methods:

• Development of a lightweight CNN architecture.
• Testing on a custom dataset of low-resolution (100x100 pixels, 96 dpi) vehicle images.
• Comparative analysis with VGG16-based CNN models regarding accuracy and computational complexity.

Main Results:

• The proposed CNN model achieved an acceptable accuracy of 92.9%.
• The model demonstrated a simpler and more lightweight solution compared to VGG16 models.
• While VGG16 models offered higher accuracy, they were more complex.

Conclusions:

• The developed CNN model provides a viable and efficient solution for vehicle classification in low-quality surveillance imagery.
• This research highlights the potential of simple, low-cost systems for enhancing intelligent transportation applications.
• Further research can leverage these findings for improved perception in intelligent systems using standard cameras.