High Resolution Tree Models: Modeling of a Forest Stand Based on Terrestrial Laser Scanning and Triangulating Scanner Data

Transcription

1 ELMF 2013, November 2013 Amsterdam, The Netherlands High Resolution Tree Models: Modeling of a Forest Stand Based on Terrestrial Laser Scanning and Triangulating Scanner Data Lothar Eysn Lothar.Eysn@geo.tuwien.ac.at With contributions from: Markus Hollaus, Milutin Milenković, Grafl Andreas, Pfeifer Norbert, Reik Leiterer Research Group Photogrammetry Department of Geodesy and Geoinformation Vienna University of Technology

2 Motivation Radiative transfer modeling ( virtual forest scenes) Synthetic models Software packages of computer graphics Often too generalized and less detail Do not follow the natural structure of vegetation Models extracted from real datasets Data from terrestrial laser scanning (TLS) Data extracted from images Manual or automatic modeling Spatially explicit tree models needed Tree model (TLS) Tree model (TLS) vs. pointcloud Requires description up to the needle scale to describe the scattering behavior high level of detail (LOD) This LOD not possible with TLS in most cases Only the main branching structure extractable Needle structure needs to be attached to tree model Fusion of models obtained from TLS data and triangulating scanner data ELMF 2013, November Amsterdam, The Netherlands 2

3 Overview Part1 Part 2 Reconstruction of wooden tree parts (stem and branches) from terrestrial laser scanning data Reconstruction of foliage / needles from triangulating scanner data Data fusion completing the tree models Final part ELMF 2013, November Amsterdam, The Netherlands 3

4 Part 1 Reconstruction of wooden tree parts (stem and branches) from terrestrial laser scanning data

5 Study Area Tharandt, Germany (South of Dresden) Old coniferous stand with little understory Homogenous, monospecies Mainly very old spruce trees with heights larger than 30 m investigated area is approximately m in size ELMF 2013, November Amsterdam, The Netherlands 5

6 TLS data acquisition Georeferencing 1 st order network GPS Survey grade GPS + Total Station GPS 3 Reference points outside forest Traject trough forest 12 planar Targets measured Free network adjustment Average deviation of planar targets (with respect to the total station data): 5 mm TS TS TLS measurements 2 nd order network TS 32 spherical targets, 12 planar targets Z+F 5006i (phase shift scanner) 34 viewpoints measured in two days point density: 16 points per 10m scan time: approx. 7 min per viewpoint In each scan approximately 6 targets visible 3D bundle block adjustment Trafo parameters The relative orientation shows an average target deviation of 4 mm GPS ELMF 2013, November Amsterdam, The Netherlands 6

7 Equirectangular Projection Equirectangular Projection TLS angle increments and measured ranging values are plotted in 2D TLS covers 360 Hz and 310 V in one scan map of x 8776 pixels (stepwidth/pixelsize is ) Maps processed for all TLS viewpoints -> Intensity or horizontal Range are visualized Different reflective properties of targets wooden tree parts are interpretable Horizontal increments r Vertical increments ELMF 2013, November Amsterdam, The Netherlands 7

8 Extracting the 3D Skeleton from 2D maps Digitizing in 2D maps -> stem and branches Easy navigation for Interpreter -> digitization is semi automatically Observation angles and range known for each point Skeleton and diameter extracted during digitizing Completing tree skeletons by digitizing in different maps Theoretically extractable diameters: Depending on scan resolution and range Scan res: 1.6 mm point 5m range 5m range; 15m range ELMF 2013, November Amsterdam, The Netherlands 8

9 Workflow tree models from TLS data Preprocessing Digitization process Volumetric models ELMF 2013, November Amsterdam, The Netherlands 9

10 Digitization process Stepwise reconstruction of the scene Animation from 20 TLS view points Transformation parameters known from registration SOCS scan1 WCS SOCS scan2 WCS etc. Old structures transformed to new viewpoint / map Gross errors become visible after transformation -> instant Check Use of human knowledge -> local data gaps can be bridged Extrusion of skeleton to cylinders by using stored radii Trees are displayed as cylinder models ELMF 2013, November Amsterdam, The Netherlands 10

11 Cylinder Model of the forest scene (II) ELMF 2013, November Amsterdam, The Netherlands 11

12 Volumetric models Validation (I) Trees modeled up to approx. ¾ of total height TLS data are limited in upper region (data gaps) Scans were only performed from the forest floor Map distortions in the upper region Polar region of the map (flattening of sphere to 2D plane) Depending on distance scanner tree Cylinders are extruded and connected correctly Multiple transformations between maps are accurate enough Upper part of intensity map ELMF 2013, November Amsterdam, The Netherlands 12

13 Volumetric models Validation (II) In total 5 randomly picked sample trees were validated Test 1: 3D deviations Stdev.: 1.0 cm to 2.0 cm Test 2: Stem validation Stdev. Of Residuals: 0.9 cm to 2.0 cm Conclusion: Trees were modeled correctly ELMF 2013, November Amsterdam, The Netherlands 13

14 Part 2 Reconstruction of foliage / needles from triangulating scanner data

15 Object Conifer branch from the study area End part of a fir (Abies alba) tree limb Consist of: one primary branch several secondary branches shoots needles Spreading approximately in one plane 80cm c.a. 80cm long c.a. 40cm side spreading from the primary branch 80cm ELMF 2013, November Amsterdam, The Netherlands 15

16 Measuring Instrument Measuring Arm (METRIS MCA, 3600 M7) Triangulating laser scanner Oriented strips in the arm coordinate system Heterogeneous resolution across strip: 0,05 mm along strip: c.a. 0,5 mm (depends on movement) Operational range up to c.a. 1,5m Operational limit limited number of points in processing ELMF 2013, November Amsterdam, The Netherlands 16

17 Measuring Acquisition method I Two-stepped scanning strategy: Primary branch scanning Separation of the secondary branches and its independent scanning Mounting Co-registration Spherical heads of pins used as targets ELMF 2013, November Amsterdam, The Netherlands 17

18 Measuring Co-registration ELMF 2013, November Amsterdam, The Netherlands 18

19 Measuring Acquired point cloud Reregistered pointcloud 13 single parts Brown part (main coord. System) > 40M points Terminology: Color indicates different scans Φ= 2,2 mm ELMF 2013, November Amsterdam, The Netherlands 19

20 Modeling Scanned part wise: Modeling branch structure Modeling shoot Cloning the shoot model Parts combined into the final branch model ELMF 2013, November Amsterdam, The Netherlands 20

21 Modeling the shoot Shoot s needles 3D Lines Shoot s stem 3D Polyline Manual digitization Topologically correct model Model Coordinate System (MCS): 2 points - starting and ending point of the 3D polyline 1 plane fitted plane trough end points of the 3D lines Reference point 2 Reference point ELMF 2013, November Amsterdam, The Netherlands 21

22 Topologically corrected 3D Polylines Manual digitization Object Coordinate System (OCS) Complex structure + needles = low accessibility point reduction procedure Modeling branch structure Point reduction: a) classified point cloud b) structure points c) skeleton Voxelization - 1 mm voxel size ~ width of the needle Local voxel density spherical kernel of 2 voxel radius Thresholding Classification to structure and needle points ELMF 2013, November Amsterdam, The Netherlands 22

23 Transformation of the shoot model: MCS (Model) OCS (Object) Cloning the shoot model (I) 2 points: starting and ending point of the segment 2 reference points of the shoot model Plane: Estimated local plane for each segment Shoot reference plane Cloning rules: End skeleton segments Whole shoot model Non-end skeleton segments Lower part of the shoot model. If skeleton segment is larger than the shoot model Lower part of the shoot model used to bridge the remaining part If skeleton segment is smaller than the lower part of the shoot model Section of lower part is used (From Reference point 1 to 2 the stations of every needle are previously derived) Based on SQL queries in a database ELMF 2013, November Amsterdam, The Netherlands 23

24 Cloning the shoot model (II) point cloud model ELMF 2013, November Amsterdam, The Netherlands 24

25 Modeled shoots Ready for cloning 3 rd year shoot 1 st year shoot 2 nd year shoot ELMF 2013, November Amsterdam, The Netherlands 25

26 Classification of the TLS point cloud (I) TLS point cloud Voxelized tree model Classified TLS point cloud Classification of the point cloud into woody and green elements Input: Reconstructed cylinder model + TLS point cloud Cylinder model and the TLS data transformed to voxel models calculation of distance model: green to woody elements calculation of the density map (for green) Thresholding of the derived products classification ELMF 2013, November Amsterdam, The Netherlands 26

27 Classification of the TLS point cloud (II) Canopy center derived Maximum of green voxel distribution in vertical extent Fill the less visible upper part in the canopy (canopy center to top of canopy): Alpha hull of the canopy based on point cloud Canopy center to top fill the hull with voxels filling randomly distributed based on a density threshold Shortest distance voxel alpha hull calculated Based on this distances the canopy is divided in different Zones classified voxel model classified TLS point cloud ELMF 2013, November Amsterdam, The Netherlands 27

28 Shoot cloning Shoot assignment for the upper crown part (1/3) based on the zones different shoots are cloned to the voxel positions Used: vector representation of the shoots LoD 1-3 years Thinout of clones shoots in the crown zone 1 50%; zone 2 35%; zone 3 20%) preserve the natural geometry / radiative transfer characteristics The determination of clone position for the lower part (2/3) is purely data driven Orientation of the shoot within each voxel: Up to the canopy center orientation to the next woody element based on the distance map (data driven orientation) If no woody element available: Elements in the inner zones are orientated vertically and towards the stem From the canopy center to the top of the canopy: orientation to the canopy center and/or based on the distance map to the next woody element Alpha hull of the voxel model of the crown Zone 1 Zone 2 Zone 3 Top crown height Canopy center Zone 2 Zone 1 Crown height 2/3 1/3 Base crown height ELMF 2013, November Amsterdam, The Netherlands 28

29 Results (I) Branches / Stem of upper crown obtained by cloning of lower branches / manual interaction ELMF 2013, November Amsterdam, The Netherlands 29

30 Results (II) 5 sample trees ELMF 2013, November Amsterdam, The Netherlands 30

31 Results (II) 5 sample trees ELMF 2013, November Amsterdam, The Netherlands 31

32 Conclusions and outlook Wooden parts of trees can be reconstructed by digitizing in 2D maps The tree models can be completed by digitizing in maps of different viewpoints The cylinder models show a high completeness and correctness for the lower part Reconstruction of upper crown regions is limited Scans from lifting platform? Other projections for the 2D maps? Modeling the shoots based on triangulating scanner data was successful Acquisition method allows independent modeling of scanned parts Shoots could be used to populate the tree models with foliage Cloning approach to simulate the foliage Cloning positions based on TLS data For larger objects, the degree of automation should be increased Automatic extraction of the skeleton needs to be investigated (2D or 3D) The method will be tested on other study areas (already started) ELMF 2013, November Amsterdam, The Netherlands 32

33 Thank you for your attention! Please read our papers: A Practical Approach for Extracting Tree Models in Forest Environments Based on Equirectangular Projections of Terrestrial Laser Scans Lothar Eysn, Norbert Pfeifer, Camillo Ressl, Markus Hollaus, Andreas Grafl, Felix Morsdorf Remote Sens. 2013, 5(11), ; doi: /rs Modeling the tree branch structure at very high resolution Milutin Milenković, Lothar Eysn, Markus Hollaus, Wilfired Karel, Norbert Pfeifer SilviLaser Conference Proceedings", (2012), Paper ID SL le@ipf.tuwien.ac.at ELMF 2013, November Amsterdam, The Netherlands 33