Motion Occlusions for Automatic Generation of Relative Depth Maps DOI

Louiza Oudni,

Carlos Vázquez, Stéphane Coulombe

et al.

Published: Sept. 7, 2018

Recovering of the depth structure a scene from monocular video content provides an important advantage in applications such as AR (placing and removing objects) or 3D-TV 3D cinema (2D-to-3D conversion). In this paper, we present automatic method to generate relative maps sequences. It relies on dynamic occlusion cue recover order objects scene. The forward backward motion analysis between each two consecutive frames allows calculation their occlusions. We estimate using modified version EpicFlow. Our modifications optical flow made it coherent forward-backward directions without compromising its performance. Thanks new feature, occlusions are simpler calculate than approaches used relevant literature. obtained allow deduction contained image. These segmentation approach which considers both color motion. Ours results show small improvement quality while adding forward/backward coherence. With respect ordering our obtains slightly better reference computationally costly step processing.

Language: Английский

Motion Occlusions for Automatic Generation of Relative Depth Maps DOI

Louiza Oudni,

Carlos Vázquez, Stéphane Coulombe

et al.

Published: Sept. 7, 2018

Recovering of the depth structure a scene from monocular video content provides an important advantage in applications such as AR (placing and removing objects) or 3D-TV 3D cinema (2D-to-3D conversion). In this paper, we present automatic method to generate relative maps sequences. It relies on dynamic occlusion cue recover order objects scene. The forward backward motion analysis between each two consecutive frames allows calculation their occlusions. We estimate using modified version EpicFlow. Our modifications optical flow made it coherent forward-backward directions without compromising its performance. Thanks new feature, occlusions are simpler calculate than approaches used relevant literature. obtained allow deduction contained image. These segmentation approach which considers both color motion. Ours results show small improvement quality while adding forward/backward coherence. With respect ordering our obtains slightly better reference computationally costly step processing.

Language: Английский

Citations

0