Recognition of trafic situations using deep neural networks

OData support
Supervisor:
Dr. Horváth Gábor
Department of Measurement and Information Systems

In this document I present my thesis work that I made in the last semester of my BSc studies at Budapest University of Technology and Economics.

Nowadasy one of the main aproaches of computer vision to build autonomous self driving cars. This challenge is not so simple, because there are so many details and questions to be solved. For example object recognition or the optimal computation frequency. In this work I focus on the object recognition part of the problem. To be more accurate, my goal is to detect persons, bikes and cars on color camera images. To solve it I will use deep neural networks, because in the last years researchers reached very high accuracy level with them. Furthermore I will present methods like data augmentation, transfer learning to hide some handicaps that I have.

The results are good, despite of my limited resources and I have chances to develop in many ways like accuracy and used computational resources. In the future I would like to use this experience to build self driving cars for Formula Stundent races.

Downloads

Please sign in to download the files of this thesis.