Forest Health Monitoring

Crown Loss Ranking

CrownLossRanking.py

Display images and boxes predicted by RTCLE and TSCLR side-by-side. Frame by frame prediction. Video can be either from youtube or sample video recorded from drones.

Adjustable variables videoOption(str): video or youtube FPVDrone3(str): any other youtube video url you want to try weights_type(str): weights type from background subtractive methods(value, GLI or kmeans) for encoder. Note, flipping is needed (-1) for kmeans and GLI speedup(int): set 1 to 3 to adjust how many frames to skip to speed up for demo (set to 1 if not demo)

Functions: labelFrameAtBottomRight: Puts a text label (eg TSCLR or RTCLE) in video frame

Object Detection Model

objectDetectionModel.py

DetectionModel class. Functions name are very straightforward.

Variational Autoencoder

VAE.py

The most important function is build_vae. pass in the weights (set using weights_type) and this function returns vae, encoder, decoder. encoder used in TSCLR

Tree Rendering from Blender

blenderTreeGen.py

Run this in Blender. Renders tree into png format. See screen recording video in blenderModels folder.

Test on Sample Video

testOnSampleVideos.py

RTCLE model on sample leaf or tree video.

Cite the following if you use this work:

@article{kocer2022vision,
  title={Vision based Crown Loss Estimation for Individual Trees With Remote Aerial Robots},
  author={Ho, Boon and Kocer, Basaran Bahadir and Kovac, Mirko},
  journal={ISPRS Journal of Photogrammetry and Remote Sensing},
  volume = {188},
  pages = {75-88},
  year = {2022},
  issn = {0924-2716},
  doi = {https://doi.org/10.1016/j.isprsjprs.2022.04.002},
  publisher={Elsevier}
}