3DTerrain-TermiteMounds

Geometric approach to filters termite mounds from Unmanned Aerial Vehicle - LiDAR data.

main script: filter_termites_CC.py

PACKAGES NEED TO BE INSTALLED: os, numpy, pandas, sklearn, matplotlib, laspy, copy, time

STEPS:

1. Load data into CloudCompare; use 'Cloth Simulation Filter'

   SETTINGS:

        filter Scene: Steep slope
        
        cloth resolution: 1.0
        
        max. iterations: 500
        
        classification threshold: 1.
   

   OUTPUTS: Save as [name].las

   EXPLANATION: ground filtering algorithm, should comprise the complete termite mounds. Under story can be included.

   https://www.cloudcompare.org/doc/wiki/index.php?title=CSF_(plugin)

2. python: keep lowest points.

   INPUT: input_file_CSF: filtered point cloud in .las format (result of the cloth simulation filter (CloudCompare)

   PARAMETERS:

        neighbours: default 30
   

   OUTPUT:low_points_knn[neighbours]_[name].las

   EXPLANATION: file 'keep_bottom_points'. Based on the X Y values, the nearest neighbours (default: 30) are identified for each point. Only the point with the lowest Z value is kept. Duplicates are removed.

3. Calculate normals in CloudCompare, can be done via python

   INPUT: low_points_knn[neighbours]_[name].las

   SETTINGS: local surface model: Quadric; radius: default 0.75 m; Orientation: Use mininum Spanning tree - knn = 6

   convert to DIP DIRECTION/DIP DEGREE.

   OUTPUT: Save as GF_[name].txt

   EXPLANATION: https://www.cloudcompare.org/doc/wiki/index.php?title=Normals%5CConvert_to_Dip_and_Dip_direction_SFs

4. python: filter based on dip direction/dip degree

   INPUT: GF_low_points_knn[neighbours]_[name].txt

   PARAMETERS:

        th_dip_degree: default (7,87) (degrees)
        
        th_NN: default 50 
        
        th_clust: default 30
        
        th_distance: default .75 m 
        
        th_RMSE: default 1
        
        extra_info: default False
   

   OUTPUT: csf_to_calculate_normals.txt

   EXPLANATION: file 'mounddetection1':

            1. data is filtered based on dip degree: all points between th_dip_degree remain
   
            2. data is filtered: points with less than th_NN points within a 2 m radius are discarded
            
            3. Data is clustered using agglomerative clustering (distance thershold = th_clust)
            
            4. extra points are added to each cluster, from the 'low_points' filtered data
            
            5. Cluster is checked if it has a conical shape: center is optimized based on dip direction. If optimized center point is > th_distance from mean (x,y), or RMSE of the optimalisation > th_RMSE, cluster is discarded.
            Creates a file 'resultsoptimizingcenter.csv' which lists the RMSE before and optimizing the termite mound center, and the distance between the initial center and optimized. Creates a folder 'plots' in which the optimized center is visualised. Creates a file 'temporaryresultsmounds.txt'. 
            
            6. extra points are added to each cluster, coming from the 'CSF' data
   

5. Calculate normals in CloudCompare, can be done via python

   INPUT: csf_to_calculate_normals.txt

   SETTINGS: local surface model: Quadric; radius: default 0.30 m; Orientation: Use mininum Spanning tree - knn = 6

   convert to DIP DIRECTION/DIP DEGREE.

   OUTPUT: Save as GF_csf_to_calculate_normals.txt

   EXPLANATION: https://www.cloudcompare.org/doc/wiki/index.php?title=Normals%5CConvert_to_Dip_and_Dip_direction_SFs

6. python: extra filter based on dip degree.

   INPUT: GF_csf_to_calculate_normals.txt

   PARAMETERS:

        th1: default 77 (%)
        
        th2: default 77 (degrees)
        
        n: default 0.15 (m)
   

   OUTPUT:

         termite_mounds.txt: labelled termite mounds
   
         termite_mounds.png: map of the termite mounds
   

   EXPLANATION: file 'mounddetection2'. all cluster of which > th1 % of points (above n m ground surface) have a dip degree > th2° are discarded.