Research seminar "An overview of multiclass detection methods"

Abstract:

A long-standing goal of computer vision has been to design a system that is able to recognize various types of objects in cluttered scenes. The examples of detecting various kinds of objects include multi-view and multi-pose object detection but also heterogeneous object detection. In the first part of talk, we provide an overview of the state of the art multiclass detection methods. The methods are compared with respect to the classifier structure and the learning method. In the second part, we discuss the applicability of the described methods in the area of traffic sign detection. As a case study, we provide an experimental evaluation of traffic sign detection using a multiplicative kernel method.

CV:

Valentina Zadrija has received the M. Eng. (dipl.ing.) degree in Computer science in  June 2008, from the Faculty of Electrical Engineering at the University of Zagreb, Croatia. From 2008 till 2010, she was employed as a research asistant at the Department of Electronics, Microelectronics, Computer and Intelligent Systems at the same Faculty. From 2008 - present, she participated in various research projects, including a project funded by the Unity through Knowledge Fund (UKF), Croatian Science Foundation and a bilateral Croatian-Austrian research project. She has co-authored six papers published in the proceedings of international conferences. Her research interests include object detection and recognition with an emphasis on the localization and mapping for navigation purposes.