Online 3D Edge Reconstruction of Wiry Structures from Monocular Image Sequences

Hyelim Choi, Minji Lee, Jiseock Kang, Dongjun Lee

Keywords: Multi-view stereo, structure from motion, edge reconstruction, monocular vision, wiry objects

Abstract: Three-dimensional (3D) reconstruction of wiry structures from vision suffers from thin geometry, lack of texture, and severe self-occlusions. We propose an online framework for reconstructing wiry structures whose skeletons are mainly straight as commonly found in man-made real-world objects in three dimensions (3D) from monocular image sequences. For an efficient and informative representation useful to address the harsh geometric nature of wiry objects (e.g., severe self-occlusion), we adopt a representation based on straight edges constructed from points. Specifically, we employ a robust maximum a posteriori (MAP) inference to construct sparse 3D points and subsequently use these sparse points to generate edge candidates whose beliefs are updated in a Bayesian manner. Then we take the set of 3D edges with beliefs greater than a threshold and apply a post-processing step to reject false edges. Experimental validation demonstrates the superior performance of our proposed framework in reconstructing 3D edges of wiry structures compared to existing state-of-the-art algorithms. We also demonstrate a manipulation task using the reconstruction that showcases the potential of the method to be easily used for subsequent robotic tasks.

Submission Number: 26