ISBN 9783843903820

978-3-8439-0382-0, Reihe Informatik

Marcel Brückner
Active Self-Calibration of Multi-Sensor Systems

192 Seiten, Dissertation Friedrich-Schiller-Universität Jena (2012), Softcover, A5

Zusammenfassung / Abstract

Several different applications already rely on multi-camera systems to concurrently acquire images from different viewpoints. Many of the cameras in daily life multi-camera systems are mounted on pan-tilt units to increase the observable area. A quite new technology are Time-of-Flight cameras which are able to simultaneously record depth and intensity images at a high frequency. The combination of these novel Time-of-Flight cameras with classical cameras to a multi-sensor system opens the doors to fascinating new applications. However, in order to be able to use the different sensor information in a beneficial way, an accurate calibration of the multi-sensor system is necessary.

In this dissertation, we develop a method for the active self-calibration of multi-camera and multi-sensor systems. Self-calibration in this context means that neither any user interaction nor artificial calibration objects are necessary. The system estimates its calibration parameters purely from the images it records from its environment. During this calibration, active control of the camera's orientation and zoom is exploited to simplify the calibration problem and reach a higher robustness and accuracy. Our calibration covers the entire intrinsic and extrinsic camera parameters as well as the hand-eye rotation between pan-tilt unit and camera.

All methods described in this dissertation are quantitatively evaluated in various experiments on simulated and real data. The results of these experiments illustrate the robustness and high accuracy of our approach. As a proof-of-concept, we use the obtained calibration to estimate the 3D body pose of a person from several 2D body pose estimates. This application gives a qualitative impression of the accuracy of the calibration obtained by our method.