Fallen dead tree detection is conducted in ALS data using contextual classification with CRFs and spectral clustering via the Ncut algorithm. A voting framework for retrieving cylindrical shapes based on kernel density estimation is applied for fallen stem mapping in TLS data. A level-set method with priors is used to segment dead tree crowns in aerial CIR imagery, and a new 3D shape descriptor is proposed for detecting dead trunks from ALS data. The semi-supervised entropy regularized logistic classifier is combined with active learning through expected error reduction. For both fallen and standing dead wood, detection accuracies of up to 80% can be reached based on experiments on data from the Bavarian Forest National Park.
«
Fallen dead tree detection is conducted in ALS data using contextual classification with CRFs and spectral clustering via the Ncut algorithm. A voting framework for retrieving cylindrical shapes based on kernel density estimation is applied for fallen stem mapping in TLS data. A level-set method with priors is used to segment dead tree crowns in aerial CIR imagery, and a new 3D shape descriptor is proposed for detecting dead trunks from ALS data. The semi-supervised entropy regularized logistic...
»