Research seminar "Stereoscopic motion estimation using multi-frame feature correspondences"

Abstract

With advancement in computer performance and camera quality, stereo vision is becoming an increasingly important source of information in traffic, robotics and other application domains. We consider applications in egomotion estimation which are also known as visual odometry. These applications simultaneously estimate the structure of the scene and the camera motion by minimizing the total reprojection error in the two image planes. This seminar provides an overview of techniques allowing us to carry out such optimization on subsequences of image pairs in real time. We especially consider suitable techniques for point feature tracking throghout multiple image frames and simultaneous optimization of structure and motion. At the end we shall present experimental results obtained on our own dataset collected from a moving vehicle as well as on the KITTI dataset.

CV

Ivan Krešo received his BSc and MSc degree from the Faculty of Electrical Engineering and Computing in Zagreb. He finished the undergraduate and master study program in Computer Science in 2011 and 2013. Currently, he is employed at the Faculty of Electrical Engineering and Computing, University of Zagreb, as a research engineer at the Department of Electronics, Microlelectronics, Computer and Intelligent Systems, where he is pursuing his PhD degree. His research interests are in the area of computer vison and machine learning.