Home Home
Introduction Introduction
Goals Goals
Our model Our model
2D results 2D results
3D results 3D results
Conclusion Conclusion
References References
Word-book Word-book
francais english
3D results 3D results Conclusion Conclusion References References

Conclusion and Acknowledgements

We proposed models to extract 2D and 3D vegetation resource parameters from high resolution aerial Colour Infrared photographs of forests. Based on a stochastic geometry approach, the models consider that each tree is one geometric object to be extracted in the image. It turns out to be efficient for segmenting the tree crowns in different kinds of stands, and gives precious information on the stand at the scale of the tree (number of trees, their size, density, ...). The 3D model for the data energy term is of particular interest because it gives access to statistics such as the wood volume. Such automatic algorithms are of economic importance because they can complete the work of forest managers and operators by giving them a more precise knowledge on the forest at a reasonable cost.

Future work will involve some tests on other kinds of images, and the implementation of a cooperation between the different models. Indeed, we could imagine a pre-processing of the image which would cut it into pieces and determine the correct model to be applied for vegetation resource extraction. Then, much work has to be carried out using shape and texture information, in order to try to associate a group of species or a class of age to one object by studying its shape, its texture, and the stand it is belonging to. The radiometry could then help to classify as much as possible the object (distinguishing between deciduous or conifer for instance).

The work of the first author has been supported by a grant from the MAS Laboratory of Ecole Centrale Paris. Part of this work has been conducted within the INRIA ARC "Mode de Vie" joint research program (Ariana Research Group and Digiplante Research Groups from INRIA, MAS Laboratory, LIAMA from Chinese Academy of Sciences, IFN). The authors would like to thank IFN for providing the data and for interesting discussions, and Digiplante Research Groupe for providing the AMAP Orchestra Software.

Guillaume Perrin - icone mail XHTML valide XHTML valide