A Cost-Effective Automatic 3D Reconstruction Pipeline for Plants Using Multi-view Images

Lu Lou, Yonghuai Liu, Minglan Sheng, Jiwan Han, John H. Doonan

Research output: Chapter in Book/Report/Conference proceedingConference Proceeding (Non-Journal item)

22 Citations (SciVal)

Abstract

Plant phenotyping involves the measurement, ideally objectively, of characteristics or traits. Traditionally, this is either limited to tedious and sparse manual measurements, often acquired destructively, or coarse image-based 2D measurements. 3D sensing technologies (3D laser scanning, structured light and digital photography) are increasingly incorporated into mass produced consumer goods and have the potential to automate the process, providing a cost-effective alternative to current commercial phenotyping platforms. We evaluate the performance, cost and practicability for plant phenotyping and present a 3D reconstruction method from multi-view images acquired with a domestic quality camera. This method consists of the following steps: (i) image acquisition using a digital camera and turntable; (ii) extraction of local invariant features and matching from overlapping image pairs; (iii) estimation of camera parameters and pose based on Structure from Motion(SFM); and (iv) employment of a patch based multi-view stereo technique to implement a dense 3D point cloud. We conclude that the proposed 3D reconstruction is a promising generalized technique for the non-destructive phenotyping of various plants during their whole growth cycles.
Original languageEnglish
Title of host publicationAdvances in Autonomous Robotics Systems
Subtitle of host publication15th Annual Conference, TAROS 2014, Birmingham, UK, September 1-3, 2014. Proceedings
EditorsMichael Mistry, Aleš Leonardis, Mark Witkowski, Chris Melhuish
PublisherSpringer Nature
Pages221-230
ISBN (Electronic)978-3-319-10401-0
ISBN (Print)978-3-319-10400-3
DOIs
Publication statusPublished - 2014

Publication series

NameLecture Notes in Computer Science
Volume8717
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Keywords

  • phenotyping
  • multi-view images
  • Structure from Motion (SfM)
  • multi-view stereo
  • 3D reconstruction

Fingerprint

Dive into the research topics of 'A Cost-Effective Automatic 3D Reconstruction Pipeline for Plants Using Multi-view Images'. Together they form a unique fingerprint.

Cite this