Full 3D Plant Reconstruction via Intrusive Acquisition

COMPUTER GRAPHICS forum

Kangxue Yin1      Hui Huang1*      Pinxin Long1     Alexei Gaissinski2     Minglun Gong3     Andrei Sharf4
1 Shenzhen VisuCA Key Lab / SIAT     2 Tel Aviv University     3 Memorial University of Newfoundland     4 Ben Gurion University

Figure 1: Reconstruct complete 3D plant models (center) with heavily self-occluded leaves (Aglaonema crispum on top) and highly-curved, non-developable leaves (Dracaena sanderiana at bottom). The texture mapped rendering results (right) closely resemble the plant photos (left).

Abstract

Digitally capturing vegetation using off-the-shelf scanners is a challenging problem. Plants typically exhibit large self-occlusions and thin structures which cannot be properly scanned. Furthermore, plants are essentially dynamic, deforming over the time, which yield additional difficulties in the scanning process. In this paper we present a novel technique for acquiring and modeling of plants and foliage. At the core of our method is an intrusive acquisition approach, which disassembles the plant into disjoint parts that can be accurately scanned and reconstructed offline. We use the reconstructed part meshes as 3D proxies for the reconstruction of the complete plant and devisea global-to-local non-rigid registration technique that preserves specific plant characteristics. Our method is tested on plants of various styles, appearances, and characteristics. Results show successful reconstructions with high accuracy with respect to the acquired data.

Overview


Figure 2: Overview of our intrusive acquisition and plant modeling algorithm.

Results

Figure 3: Reconstruction of four complex plants. From top to bottom, we have Stromanthe sanguinea, Calathea majestica, Aglaonema modestum, and Lilium casa blanca. The results show that the presented global-to-local registration step (d) can effectively reduce the reconstruction error caused by the initial imprecise positioning (b) of the leaf models. The color of scan points in (b) and (d) encodes the closest distance to the meshes (red is further and blue is closer). For better illustration, the effect of registration is also demonstrated using selected 2D cross sections in (c) and (e). The texture (from plant photos) mapped models in (d) are rendered in (f), which faithfully resemble the plant’s appearances in the real photos (insets).


Acknowledgments

We would like to thank the anonymous reviewers for their constructive comments. This work was supported in part by NSFC (61379090, 61202224, 61202221), National 973 Program (2014CB360503), Shenzhen Innovation Program (CXB201104220029A, JCYJ20140901003939034, JCYJ20140901003938994), SIAT Innovation Program for Excellent Young Researchers (201305, 201402),NSERC(293127) and the Israel Science Foundation.

BibTex

@ARTICLE{yin2015Plant,
    title={Full 3D Plant Reconstruction via Intrusive Acquisition},
    author={Yin, Kangxue and Huang, Hui and Long, Pinxin and Gaissinski, Alexei and Gong, Minglun and Sharf, Andrei},
    journal={Computer Graphics Forum},
    volume = {34},
     issue = {5},
    year={2015},
}

Copyright © 2016-2018 Visual Computing Research Center