WE2.TO9.2.pptx

301 views
287 views

Published on

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
301
On SlideShare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
1
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

WE2.TO9.2.pptx

  1. 1. ADAPTIVE MORPHOLOGICAL FILTERING FOR DEM GENERATION<br />KyoHyouk Kim and Jie Shan<br />Geomatics Engineering<br />School of Civil Engineering<br />Purdue University <br />July 27th 2011<br />IGARSS 2011 Annual Conference<br />
  2. 2. OUTLINE<br />Introduction<br />Previous Works<br />Proposed Approach<br />Tests<br />Summary<br />2/28<br />
  3. 3. OUTLINE<br />Introduction<br />LiDAR<br />Filtering<br />Previous Works<br />Proposed Approach<br />Tests<br />Summary<br />3/28<br />
  4. 4. INTRODUCTION<br />LiDAR (Light Detect And Ranging)<br />One of remote sensing technologies providing point clouds with highly accurate collections of (X,Y,Z) at observed points.<br />Vertical accuracy : 6 to 30 cm<br />Horizontal accuracy : 10 to 46 cm<br />Used as one of primary data sources to applications:<br />DEM (Digital Elevation Model)<br />Forest mapping <br />Building footprints & 3D building reconstruction<br />4/28<br />
  5. 5. INTRODUCTION<br />Filtering<br />Separation between ground points and nonground points.<br />Ground points : Digital Elevation Model (DEM).<br />Non-ground points : Building models, forest structure, canopy mapping<br />In nearly all applications, filtering should be done first.<br />Various filtering algorithms have been proposed.<br />Slope-based filters<br />Linear prediction filters<br />Clustering (or segmentation) filters<br />Morphological filters<br />5/28<br />
  6. 6. INTRODUCTION<br />Example<br />6/28<br />
  7. 7. OUTLINE<br />Introduction<br />Previous Works<br />General rule<br />Morphological filtering algorithm<br />Known issues & weaknesses<br />Proposed Approach<br />Tests<br />Summary<br />7/28<br />
  8. 8. PREVIOUS WORKS<br />General rule<br />What is ground surface ?<br />Ground surfaces are relatively smooth & continuous<br />Locally spanned through the lowest parts.<br />EX: discontinuous ground surface and negative blunders.<br />All filtering algorithms try to<br />Identify points on the continuous and smooth ground surface<br />Remove points lying on non-ground objects<br />Measures of discontinuity<br />Abrupt slope or elevation difference<br />8/28<br />
  9. 9. PREVIOUS WORKS<br /> Morphological filtering<br />Commonly used in the image processing fields to :<br />Remove noise<br />Enhance images<br />Extract features<br />Published papers for LiDAR filtering<br />Kilian, Halla, and Englich (1996), Kilian et al. (1996), Zhang et al. (2003), Zaksek and Pfeifer (2006), Zhang and Cui (2007), Chen et al. (2007,2009). <br />9/28<br />
  10. 10. PREVIOUS WORKS<br /> Principles<br />Morphological opening operation : Erosion followed by dilation<br />Erosion : <br />Dilation :<br />Objects with different size<br />Apply different window size <br />If terrain is not flat<br />use maximum slope<br />d0<br /><br />d<br />10/28<br />
  11. 11. PREVIOUS WORKS<br />Known issues & weaknesses<br />Parameters have a great impact on the filtering result<br />Slope : allowed minimum elevation difference<br />Window size : objects with various sizes<br />Small window : Can not remove larger buildings<br />Large window : Highly possible to flatten ground surface<br />No optimal set of parameters supporting various types of terrain<br />Require a priori information about the target area.<br />Require trial-and-error process to find the best set of parameters<br />11/28<br />
  12. 12. OUTLINE<br />Introduction<br />Previous Works<br />Proposed Approach<br />Adaptive morphological filtering<br />Tests<br />Summary<br />12/28<br />
  13. 13. PROPOSED APPROACH<br />Adaptive morphological filtering<br />Use of regular grid with grid size g ( average point spacing)<br />Cell with multiple points : point with the lowest elevation<br />Empty cell : elevation of the nearest point<br />Keeps the indices of original LiDAR points<br />1D (or 2D) Morphological erosion operation (W = 3).<br />Filtering is applied only to points of discontinuity iteratively.<br />Relevant parameters are adaptively adjusted in each iteration<br />13/28<br />
  14. 14. PROPOSED APPROACH<br />Use of 2D regular grid<br />14/28<br />
  15. 15. PROPOSED APPROACH<br />Discontinuity Measure<br />Identify points lying on the edge between ground and objects<br />Use of nominal value for thresholds (No fixed thresholds)<br />Ex: 50, 75 or 90 percentile<br />Threshold is determined before each iteration<br />Residual<br />Height difference<br />Slope<br />15/28<br />
  16. 16. PROPOSED APPROACH<br />Example<br />One LiDAR profile<br />Slope<br />Residual<br />16/28<br />
  17. 17. PROPOSED APPROACH<br />Workflow<br />For i=1 to # of rows (or # of columns)<br />Determine residuals (ri) of all points<br />Ifri> N percentile of r (N=50,75,90)<br />If hi - (New hi) > Hmin<br />hi = New hi<br />Until # of filtered<br /> points = 0<br />End<br />17/28<br />
  18. 18. PROPOSED APPROACH<br />Example<br />70th row<br /> - 50 percentile<br /> - Hmin = 1.0m<br />Purdue campus<br />18/28<br />
  19. 19. PROPOSED APPROACH<br />Refinement<br />Type I (omission) error : remove ground points mistakenly<br />Caused by (1) negative blunders and (2) discontinuous ground<br />Type II (commission) error : classify objects to ground points<br />Caused by (1) nonground points with smaller elevation than Hmin<br />19/28<br />
  20. 20. PROPOSED APPROACH<br />Example ( Type I error)<br />20/28<br />
  21. 21. PROPOSED APPROACH<br />Resolve Type I error<br />Identify continuous segments of ground points<br />Distance of any non-ground points to the line (Si(n) – Si+1(1))<br />Repeated until no more ground points are restored<br />d<br />21/28<br />
  22. 22. OUTLINE<br />Introduction<br />Previous Works<br />Proposed Approach<br />Tests<br />Summary of data sets<br />Evaluations<br />Summary<br />22/28<br />
  23. 23. INITIAL RESULTS<br />Test data sets<br /><ul><li>High relief (mixed slope)
  24. 24. Small step-wise features
  25. 25. Discontinuous surface
  26. 26. Almost flat
  27. 27. Buildings mixed with trees
  28. 28. High relief (mixed slope)
  29. 29. Dense vegetation</li></ul>23/28<br />23/26<br />
  30. 30. INITIAL RESULTS<br />Filtered ground surface (1)<br /><ul><li>Hmin= 1.5m
  31. 31. Residual threshold = 50 percentile</li></ul>24/28<br />
  32. 32. INITIAL RESULTS<br />Filtered ground surface (2)<br /><ul><li>Hmin = 1.0 m
  33. 33. Residual threshold = 50 percentile</li></ul>25/28<br />
  34. 34. INITIAL RESULTS<br />Filtered ground surface (3)<br /><ul><li>Hmin = 1.5 m
  35. 35. Residual threshold = 50 percentile</li></ul>26/28<br />
  36. 36. OUTLINE<br />Introduction<br />Previous Works<br />Proposed Approach<br />Tests<br />Summary<br />27/28<br />
  37. 37. SUMMARY<br />Adaptive morphological filtering<br />Minimize the effects of parameters<br />Filtering is applied only to points on the boundary iteratively<br />Residual is used as the measure id discontinuity<br />Residual threshold is adaptively adjusted with nominal value<br />Promising results from various terrain conditions, but :<br />Type I error is often significant in case of discontinuous terrain<br />Need more robust stopping criterion<br />28/28<br />

×