Dynamic horizontal image translation in stereo 3D
Loading...
Date
2017
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Im Bereich Stereo 3D (S3D) bezeichnet „Dynamic Horizontal Image Translation
(DHIT)“ das Prinzip, die S3D-Ansichten einer Szene horizontal in entgegengesetzte
Richtungen zu verschieben, wodurch die dargestellte Szene in der Tiefe
verschoben wird. Dies wird vor allem im Kontext von „Active Depth Cuts“
eingesetzt. Hier werden die S3D-Ansichten vor und nach einem Szenenschnitt
so verschoben, dass es nicht zu starken, störenden Tiefensprüngen kommt.
Die menschliche Wahrnehmung der DHIT wurde experimentell untersucht.
Eine der wichtigsten Erkenntnisse war, dass es starke individuelle Unterschiede
in der Empfindlichkeit gegenüber der DHIT gibt. Daher wird empfohlen die
Verschiebungsgeschwindigkeit einer S3D-Ansicht nicht höher als 0,10 °/s bis
0,12 °/s zu wählen, sodass Zuschauerinnen und Zuschauer nicht von der DHIT
gestört werden.
Bei der DHIT kommt es zu einer Verzerrung der dargestellten Szenentiefe. Dies
wird bei dem vorgeschlagenen Ansatz „Distortion-Free Dynamic Horizontal
Image Translation (DHIT+)“ kompensiert, indem der Abstand zwischen den
S3D-Kameras durch Verfahren der Ansichtensynthese angepasst wird. Dieser
Ansatz zeigte sich signifikant weniger störend im Vergleich zur DHIT. Die
Ansichten konnten ohne Wahrnehmungsbeeinträchtigung etwa 50% schneller
verschoben werden.
Ein weiteres vorgeschlagenes Verfahren ist „Gaze Adaptive Convergence in
Stereo 3D Applications (GACS3D)“. Unter Verwendung eines Eyetrackers wird
die Disparität des geschätzten Blickpunkts langsam über die DHIT reduziert.
Dies soll die Ermüdung des visuellen Systems mindern, da die Diskrepanz zwischen
Akkommodation und Konvergenz reduziert wird. In einem Experiment
mit emuliertem Eye-Tracking war GACS3D signifikant weniger störend als eine
normale DHIT. Im Vergleich zwischen dem kompletten GACS3D-Prototypen
und einer Bildsequenz ohne jegliche Verschiebungen konnte jedoch kein
signifikanter Effekt auf den subjektiven Betrachterkomfort registriert werden. Eine
Langzeituntersuchung der Ermüdung des visuellen Systems ist nötig, was über
den Rahmen dieser Dissertation hinausgeht. Da für GACS3D eine hochgenaue
Schätzung der Blickpunktdisparität benötigt wird, wurde die „Probabilistic
Visual Focus Disparity Estimation“ entwickelt. Bei diesem Ansatz wird die
3D-Szenenstruktur in Echtzeit geschätzt und dazu verwendet, die Schätzung
der Blickpunktdisparität deutlich zu verbessern.
Dynamic horizontal image translation (DHIT) denotes the act of dynamically shifting the stereo 3D (S3D) views of a scene in opposite directions so that the portrayed scene is moved along the depth axis. This technique is predominantly used in the context of active depth cuts, where the shifting occurs just before and after a shot cut in order to mitigate depth discontinuities that would otherwise induce visual fatigue. The perception of the DHIT was investigated in an experiment. An important finding was that there are strong individual differences in the sensitivity towards DHIT. It is therefore recommended to keep the shift speed applied to each S3D view in the range of 0.10 °/s to 0.12 °/s so that nobody in the audience gets annoyed by this approach. When a DHIT is performed, the presented scene depth is distorted, i.e., compressed or stretched. A distortion-free dynamic horizontal image translation (DHIT+) is proposed that mitigates these distortions by adjusting the distance between the S3D cameras through depth-image-based rendering techniques. This approach proved to be significantly less annoying. The views could be shifted about 50% faster without perceptual side effects. Another proposed approach is called gaze adaptive convergence in stereo 3D applications (GACS3D). An eye tracker is used to estimate the visual focus whose disparity is then slowly reduced using the DHIT. This is supposed to lessen visual fatigue since the infamous accommodation vergence discrepancy is reduced. GACS3D with emulated eye tracking proved to be significantly less annoying than a regular DHIT. In a comparison between the complete prototype and a static horizontal image translation, no significant effect on subjective visual discomfort could be observed, however. A long-term evaluation of visual fatigue is necessary, which is beyond the scope of this work. In GACS3D, highly accurate visual focus disparity is required. Therefore, the probabilistic visual focus disparity estimation (PVFDE) was developed, which utilizes a real-time estimation of the 3D scene structure to improve the accuracy by orders of magnitude compared to commonly used approaches.
Dynamic horizontal image translation (DHIT) denotes the act of dynamically shifting the stereo 3D (S3D) views of a scene in opposite directions so that the portrayed scene is moved along the depth axis. This technique is predominantly used in the context of active depth cuts, where the shifting occurs just before and after a shot cut in order to mitigate depth discontinuities that would otherwise induce visual fatigue. The perception of the DHIT was investigated in an experiment. An important finding was that there are strong individual differences in the sensitivity towards DHIT. It is therefore recommended to keep the shift speed applied to each S3D view in the range of 0.10 °/s to 0.12 °/s so that nobody in the audience gets annoyed by this approach. When a DHIT is performed, the presented scene depth is distorted, i.e., compressed or stretched. A distortion-free dynamic horizontal image translation (DHIT+) is proposed that mitigates these distortions by adjusting the distance between the S3D cameras through depth-image-based rendering techniques. This approach proved to be significantly less annoying. The views could be shifted about 50% faster without perceptual side effects. Another proposed approach is called gaze adaptive convergence in stereo 3D applications (GACS3D). An eye tracker is used to estimate the visual focus whose disparity is then slowly reduced using the DHIT. This is supposed to lessen visual fatigue since the infamous accommodation vergence discrepancy is reduced. GACS3D with emulated eye tracking proved to be significantly less annoying than a regular DHIT. In a comparison between the complete prototype and a static horizontal image translation, no significant effect on subjective visual discomfort could be observed, however. A long-term evaluation of visual fatigue is necessary, which is beyond the scope of this work. In GACS3D, highly accurate visual focus disparity is required. Therefore, the probabilistic visual focus disparity estimation (PVFDE) was developed, which utilizes a real-time estimation of the 3D scene structure to improve the accuracy by orders of magnitude compared to commonly used approaches.
Description
Table of contents
Keywords
Stereo 3D, Dynamic Horizontal Image Translation, Visual Perception, 3D Visual Focus Estimation, View Synthesis