Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Detection of pose changes for spatial objects from projective images

: Selby, Peter; Sakas, Georgios; Walter, Stefan; Groch, W.-D.; Stilla, Uwe

Stilla, U. ; International Society for Photogrammetry and Remote Sensing -ISPRS-; TU München, Institut für Photogrammetrie und Kartographie:
PIA07, Photogrammetric Image Analysis. Part A : Papers Accepted on the Basis of Peer-Reviewed Full Manuscripts, 19-21 September 2007, Munich, Germany
München: Technische Universität München, Institut für Photogrammetrie und Kartographie, 2007 (The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 36, Part 3/W49A)
ISSN: 1682-1750
Conference "Photogrammetric Image Analysis" (PIA) <2007, München>
Fraunhofer IGD ()
pose estimation; image registration; landmark; spatial objects

Numerous fields of application effort the detection of pose changes for 3 dimensional objects in six degrees of freedom (6 DoF). Automatic procedures that exploit 2D images for the detection of pose changes can be used for example for tracking object movements, for quality control or for the verification of the alignment of patients in radiation treatment devices. In this contribution we present two different solutions for the detection of pose changes that base on the comparison of two 2D images resulting from the projection of an object in the new pose and a 3D volume of the same object in a known reference alignment.
Whereas for the first solution we use an object where we can clearly extract landmarks useable as reference positions for the determination of the object's alignment, we provide a second solution for objects where these landmarks cannot be extracted, which is involved automatically if necessary. In this case grey value based pose estimation is conducted by registering the computationally projected reference 3D volume to the 2D images. As reference data for the object with known alignment, CT slices will be used, as they are provided for the alignment of patients in radiation treatment devices. Two X-ray images of the same object in an unknown pose can then be compared to the reference data to determine the respective pose change, which may consist of 3 rotations and 3 translations. Using both approaches to determine patient misalignments in treatment devices shows, that both methods result in highly accurate pose detections and that the second method, despite being less accurate and more time consuming, is an appropriate solution in cases where landmark detection fails.