Advanced vision based vehicle classification for traffic surveillance system using neural networks

Abstract

This master thesis focus on traffic monitoring, which are of importance to fulfill planning and traffic management of road networks. An important requirement is data interpretation accuracy to provide adequate characteristic data from the acquired vision-data. A vision-based system has been developed, using new methods and technologies to achieve an automated traffic monitoring system, without the use of additional sensors.

The thesis is based upon Erik Sudland’s master thesis from 2016, which investigated available litterateur containing adequate algorithms for traffic monitoring. However in the current master thesis, methods have been further analyzed and experimentally optimized on vision-data from real traffic situations. In addition, a new classification method based upon neural networks has been implemented and verified with successful results