Accurate and useful models of plants in three dimensions, based on multiple colour images using a low-cost camera
Our reconstruction work aims to produce accurate and useful models of plants in three dimensions, based on multiple colour images using a low-cost camera. Beginning with an initial point cloud, obtained using standard structure from motion tools, the scene is first filtered, and segmented into small patches that represent sections of leaf surface. Each surface is individually represented by a best-fit plane, and the boundary of each surface is adjusted to better fit the image and modelling data. This adjustment is achieved using the level set method, which optimises the smoothness of each surface, while growing or shrinking them in response to image information and neighbouring surfaces.
Our work has been published in a paper in Plant Physiology, and the technical methods are published in the CVPPP workshop at ECCV 2014. The software is open source, and can be downloaded from our code repository. The compiled executables can be downloaded here.
Quick Start Guide
A quick start guide is available to provide a brief introduction to the use of the software. The guide overviews the reconstruction of the rice dataset that was shown in the Plant physiology publication. This dataset can be downloaded here.
The software is run using the Windows command line, and requires that specific files (such as images and calibration files) are supplied in the correct location. These details, along with input parameters can be found in the documentation.
This project is currently under development. Among other possibilities, we are working on updating the tool to support lidar and other laser data.
For further information on the software, please contact Michael Pound.
Associated papers are available in Plant Physiology and CVPPP 2014 (Coming soon).