3D-reconstruction of rigid constructions from the monocular video

Manshin Ilya Mikhailovich, Falkov Georgy Aleksandrovich, Zuev Sergei Valentinovich

To date, approaches based on template representations of a particular class have achieved some success. The disadvantage of such methods is the lack of flexibility of use in relation to unknown categories of objects. The paper presents a template-free approach for studying 3D shapes in one video. It uses an analysis-by-synthesis strategy that allows you to visualize the silhouette of an object for comparison with video surveillance. Without relying on a category-specific form template, the method reconstructs rigid 3D structures from videos of unknown classes.

Keywords:3D shape restoration, monocular video, mash, rest shape, rendering, 3D shape.