Tracking and Reconstruction in a Combined Optimization Approach
Olaf Kähler
University of Jena, Germany
AG4 Seminar
Olaf Kähler recieved the Diploma degree in computer science from the University of Erlangen in 2004. Since 2005 he is working as a PhD student at the Chair for Computer Vision in Jena, supervised by Prof. Denzler. His research interests focus on 3d computer vision, structure-from-motion and especially the link between tracking and reconstruction.
The structure-from-motion problem is a central task in computer vision. Most solutions follow a two step strategy: matching features between images in a first step and then extracting 3d information in a distinct, second step. However, the tracking process can greatly benefit from knowledge about the 3d scene. Also modelling the uncertainty of the tracker in the reconstruction process requires a close connection of the two steps. We address these two issues by formulating a combined approach to tracking and reconstruction.
In this formulation, an implicit feedback of 3d information to the tracking process is achieved. Also, no assumptions about the error distribution of the tracker are needed. The results are statistically optimal in case of Gaussian noise on the measured intensity values. Experiments verify an improved tracking robustness and a higher reconstruction accuracy. The approach is suited for online reconstruction and has a close to real-time performance with 5-10 fps on current computing hardware.