Automatic Individual Tree Crown Delineation

Marker-controlled watershed algorithm is very simple and effective. LiDAR point cloud is converted into 2D canopy height model (raster model) and if you can imagine inverting the raster model. The depression of the surface model can be identified as tree top and the edge of the derived watershed boundary can be identified as tree crown boundary. The biggest challenges for this method is to overcome the assumption that one tree hold a single tree top and the fine the proper “scale” for segmenting the watershed. It’s sometimes managed by train-and-error approach for finding those parameters. I’m currently looking for ways to overcome these challenges.