Contest No. 2

Individual tree detection from Unmanned Aerial Vehicle (UAV)

Description:

Data captured by the unmanned aerial vehicles (UAVs) have been recently considered as promising sources of structural and spectral information in the precision forestry. Detection of individual tree is an important attribute that is essential for forest inventory and management. In this contest, the capability of products obtained by small and cost-effective UAVs will be evaluated to detect individual broadleaf trees within a portion of low-land Hyrcanian forests of Iran (Fig 1). Accordingly, the individual steps of the data processing could be as follows:

a-      Preprocessing of point cloud: including cropping, noise reduction, filtering, and point filtering.

b-      Visual geometric assessment between point clouds and the resulted ortho-mosaic.

c-      Segmentation and extraction of individual trees based on joint or sole use of point clouds and ortho-mosaic

d-      Extraction of a single trees for each tree segment, exported in a common vector format (shape files).

e-      Representation of the results as shape file format with a specific number for each tree plus a geometric location (x,y)

f-       Writing a technical report by describing all operational individual steps

Fig 1: The study area, a low-land Hyrcanian forests of Iran.

Dataset:

To implement the methodology two main datasets are available:

-          Ortho-mosaic from an RGB sensor on-board of a Phantom 4 UAV in 25 cm spatial resolution (tif format)

-          Point clouds generated from UAV in a las format.

Deliverables:

After producing the final model, the following materials should be provided and send it to the email address announced by the conference in zip format:

-          Technical report in docx or pdf describing all detail analysis (NOT more than 5000 words)

-          Produced model in shape format (.shp): the shape file should be opened in any standard GIS/Photogrammetry software.

-          Programming codes (if employed)

Evaluation Criteria:

The following criteria are considered for the evaluation of the accomplished tasks:

-          Overall accuracy of the detected trees

-          Commission error of tree top detection

-          Omission error of tree top detection

-          Geometric accuracy of the extracted trees.

-          To develop and use your own program code (preferably open source) instead of using existing commercial tools.

-          A comprehensive technical report (in Persian or English).

-          Level of scientific innovation.

Data availability:

The web link for downloading the point cloud data will be available upon the completion of the registration form. 

 Achnowledgement

The scientific committee would like to thank Dr. Hormoz Sohrabi and Sima Sadeghi (Tarbiat Modares University, Faculty of Natural resources, Departement of Forest Science) for acquiring and providing UAV images.