Fallen Tree Detection

This repository contains the source code for the ALS-based fallen tree detection algorithm developed by Heinaro et al. 2021. The algorithm is used via the function findFallenTrees() that is located in main\findFallenTrees.m.

Check the function documentation for a more detailed description of the inputs, outputs and usage of the function.

NOTE: The script startup.m must be called before running the function, as it adds all the required filepaths to the MATLAB path.

Algorithm

The fallen tree detection algorithm consists of four steps:

1. Reading and preprocessing the data
2. Filtering the point cloud using classification based on connected component analysis
3. Detecting fallen trees using iterative Hough transform -based line detection
4. Removing false fallen tree segments using a convolutional neural network

Steps 2 and 4 are optional. By default, the function findFallenTrees() performs step 2 using a default classifier that can be found in the folder trained_classifiers\, whereas step 4 is not performed. Steps 2 and 4 can use user-defined classifiers trained using the functions found in the folders connected_component_training\ (step 2) and final_classifier_training\ (step 4).

Example

A short example of the function can be found under example\exampleUsage.m. It contains all necessary steps, including the call to startup.m. Paths are relative, execute the script from it's folder.

Prerequisites / Installation

The repository contains all files to run a basic example or use the code on your own data.

MATLAB is required with the following add-ons (the last 3 are optional):

Name	Version	PN	Certain
MATLAB	9.7	1	true
System Identification Toolbox	9.11	4	true
Signal Processing Toolbox	8.3	8	true
Mapping Toolbox	4.9	11	true
Deep Learning Toolbox	13.0	12	true
Image Processing Toolbox	11.0	17	true
Statistics and Machine Learning Toolbox	11.6	19	true
Parallel Computing Toolbox	7.1	80	false
MATLAB Parallel Server	7.1	94	false
Polyspace Bug Finder	3.1	164	false

You can check whether all prerequisites are met by running the following code from the base directory of the repository:

files = dir('**\*.m');
fullfile = strcat( extractfield(files,'folder')', {'\'}, extractfield(files,'name')');
[requiredFiles, requiredProducts] = matlab.codetools.requiredFilesAndProducts(fullfile);

requiredProducts will contain a complete list of the above mentioned toolboxes if they are installed (missing add-ons won't be listed by matlab.codetools.requiredFilesAndProducts(fullfile) and may cause unexpected error when running this code).

Due to it's extensive use of MATLAB Toolboxes, the script is unlikely to run out-of-the-box in Octave environments.