Simultane Tiefen- und Flussbestimmung pflanzlicher Oberflächen

TitleSimultane Tiefen- und Flussbestimmung pflanzlicher Oberflächen
Publication TypePhD Thesis
Year of Publication2004
AuthorsKüsters, R
UniversityIWR, Fakultät für Physik und Astronomie, Univ.\ Heidelberg

The subject of this thesis is the stereo based 3d survey of deformable objects. This includes the calculation of spatial structure and deformations of plant surfaces. The position in space and the movement field are simultaneously estimated as depth and optical flow in multi camera image sequences. This is realized by a near baseline stereo approach. Temporal multi camera sequences are taken as a 4d data set. A linear model is used to calculate depth. The brightness change constraint equation (BCCE) is extended by disparity terms. Parameters are estimated with a local differential total least squares method, the structure tensor approach, simultaneously yielding depth and flow information. An additional extension of the BCCE allows the simultaneous estimation of flow divergence and thus depth motion. The accuracy of this techniques is quantified on synthetic and real sequences. The results show the typical behavior for the structure tensor approach, high noise stability and accuracy. As a botanical application, a method for measuring of local relative area changes of moving curved surfaces is developed. The temporal course of those growth rate measurements shows a clear diurnal rhythm. Limiting the evaluations to those of static multi camera sequences allows the 3d survey of tree canopies as a smoothed envelope. To accommodate this for extended populations, a method is developed which creates a fusion of partial 3d reconstructions. This is applied in the high resolution reconstruction of the rainforest canopy in the Biosphere 2 Center, Arizona.

Citation Keykuesters2004