Detecting and classification of vehicles using neural networks

OData support
Supervisor:
Dr. Kővári Bence András
Department of Automation and Applied Informatics

Vehicle recognition systems have been used for several tasks in the industrial environment, such as traffic counting, self-driving cars and vehicle tracking. In the specific field of object recognition, a demand has emerged for a system, which can solve online detection and classification problems. I present a solution using deep neural networks.

This paper summarizes the entire network construction process from the database building to the completely implemented neural networks based vehicle detection systems. I introduce the database building workflow and the data manipulation tools, wherewith the labelling process has improved. I also implemented neural networks used for classification in Matlab, to automate classification error correction. Finally, I present the modifications of two public, Caffe based object detection systems, by which the trained neural networks became suitable for fast and accurate vehicle detection.

The detection and classification results can be used to create a highly reliable vehicle tracking system. Thereby the full automation of the traffic counting becomes achievable.

Downloads

Please sign in to download the files of this thesis.