Star fruit

Abstract

This project marks the start of a new collaboration between computer scientists fromUniversité Grenoble Alpes biophysicists from Paris Diderot University. The objectives are to develophardware and software to help track points on a star fruit leaf as it grows using a multi-camera system and to measure their movement in 3D.

Coordinators

Franck Hétroy-Wheeler (LJK/Inria)

Julien Derr (Laboratory of Complex Matter and Systems, Paris)

Scientific findings

A camera system was designed and installed in Grenoble and Paris. In Grenoble, the Kinovis platform was customized—both in terms of hardware and software—for small volumes, high spatial resolutions, and very long time scales, as is the case in our context. The 8 (Grenoble) or 10 (Paris) cameras surrounding the factory are synchronized using a trigger and an Arduino system (Grenoble) or a USB system (Paris) programmed to take photos every k minutes, where k is a user-defined parameter. Calibration is performed using a checkerboard. For each series of 8 or 10 images, a 3D mesh model is reconstructed usingthe EPVH algorithm. A Reeb graph is then calculated and incorporated into each mesh to define a one-dimensional spatiotemporal model for analyzing the movement of the leaf stalk. Dense temporal tracking of the leaf based on feature points is a work in progress. Such tracking will allow for a more detailed analysis of movements, if possible at a scale of 10 to 100 microns.

The final stages of the project involve tracking the plant’s movement over time and measuring that movement precisely. This should be completed by July 2017. We then plan to expand the project to other plants with interesting dynamics, such as the avocado (Persea americana). A collaboration with biologists from INRA in Avignon, Clermont-Ferrand, and Montpellier is also planned (ANR grant application submitted) to apply the same type of tools on a larger scale and in the field for fruit trees and agrosystems. Finally, a new collaboration has recently begun with Florence Bertails-Descoubes (Inria Grenoble) to use the calculated, temporally consistent 3D plant models for inverse mechanical modeling.