Detecting Buildings in AHN2 LiDAR data with ArcGIS - Grontmij
3 likes1,272 views
The document discusses the update of object heights in rural urban areas using AHN2 data and LIDAR technology, detailing pricing information for LIDAR products and processing challenges. It highlights the object detection process and its results, showing a high rate of detection but also some misclassifications. Lessons learned emphasize the complexity and time-consuming nature of processing AHN2 data along with its limitations regarding completeness and correctness.
Detecting Buildings in AHN2 LiDAR data with ArcGIS - Grontmij
- 1. 1Extract buildings from AHN2 dataXander BakkerGrontmijTeam GIS
- 2. 2Some background informationClient calculates the locations for their communication network
- 3. A ‘Clutter’ holds base data on surface including object heights classification
- 4. Grontmij has provided the clutter dataset in 2004
- 5. Some updates on object heights were done in 20053Some background informationClient requested us to do an update of the object heights for urban areas currently classified as rural
- 6. Using various datasets; NKNL (plans for urban expansion), W&B (population data), aerial photography (Bing) urban expansion areas were identified
- 7. Previous updates were based on AHN1 (5m Min and Max grids)
- 8. AHN1 is outdated, AHN2 is available for large parts of interest area.4LiDAR (Light Detection And Ranging)Laser beam is sent to the earth surface
- 9. Return time defines surface height
- 11. AHN2: 30 points per m²
- 12. AHN2: precision is approx. 5 cm5LiDAR 3D point clouds
- 13. 6AHN2 productsPricing / km² (< 1.250 km²)€ 56,-€ 56,-€ 28,-€ 28,-€ 28,-€ 12,-Pricing / km² (<= 15.000 km²)€ 28,-€ 28,-€ 14,-€ 14,-€ 14,-€ 6,-Pricing / NL (> 15.000 km²)€ 455.000,-€ 455.000,-€ 227.500,-€ 227.500,-€ 227.500,-€ 97.500,-Data containsBare EarthBare Earth + ObjectsBare EarthBare EarthBare Earth + ObjectsBare EarthProductFiltered laserpointsUnfiltered laserpoints0.5m grid filtered0.5m grid filtered + interpolated0.5m grid unfiltered5m grid filteredThe purpose of AHN2 is to get a highly detailed representation of the bare earth (maaiveld). Objects (houses, trees, etc.) are filtered.
- 14. 7LASER pointsUnfiltered pointsFiltered pointsPoint cloudTIN
- 15. 8Filtered grid 0.5 m/pix (Bare Earth)Not InterpolatedInterpolated
- 16. 9Unfiltered grid 0.5m/pixThe only raster AHN2 product that holds information on objects…
- 17. 10Profile unfiltered grid 0.5m/pixditchrivertreeroadhedgefarm
- 19. 12LIDAR Analyst v4.2Vector output (3D)
- 20. Attributed data
- 21. Extraction of buildings, trees and forest
- 23. Batch processing
- 25. Parameters vary per area
- 26. Long processing time (estimated 20 minutes CPU per AHN2 subunit)
- 28. Expensive extension (>10K US$)13ArcGIS 9.4 and AHN210ArcGIS 9.3ArcGIS 9.4 10SOMA (Single Output Map Algebra) will be implemented in ArcGIS 10…
- 29. 14ArcGIS 9.3 Spatial Analyst / ArcObjectsSimple Case : Buildings exist in TOP10TOP10 buildingsRegion groupZonal maxZonal max – Focal minAHN2 rasterFocal min
- 30. 15Common ExceptionsBuildings not in TOP10
- 31. Buildings contain mainly NoData
- 32. Buildings contain extreme low values (sometimes lower than 20m below bare earth)16Detect Buildings in LiDAR dataLiturature suggests ‘Thresholding’Height
- 33. Size
- 34. Gradient
- 35. Slope17Recent investigation other organizationsObject detection using LiDAR and Satellite ImageryMin/Max
- 36. NDVIOnly works for small areas!
- 37. 18Correct AHNCorrection of extreme valuesAHN2 (DSM)SetNull DSM < BE-1Bare Earth (BE)FocalMean on NoData in DSM
- 38. 19The object detection process (1)Threshold slope and heightSlope < 55°SlopeDSMOH > 3mBare Earth (BE)Object heightsObjects (1)
- 39. 20The object detection process (2)Include objects from NoData areasRegion Group (ND3)Expand (ND2)NoData areas (ND1)DSMWater (WA0)Zonal Sum Water in NoData (WA1)
- 40. 21The object detection process (3)Include objects from NoData areasThreshold heightWater in NoData (WA1)SetNull (WA2)Zonal Mean (ND4)DSMFinal ObjectsObjects previous steps
- 42. 23ResultsBuildings not in TOP10 are detected
- 43. Buildings mainly covered with NoData are depicted as well1before231after23
- 45. Processing time for 1 subunit is 3 min.
- 46. Batchprocessing
- 47. Good alternative for lacking data in TOP10
- 48. Highways need to be filtered out
- 49. Some buildings not detected
- 50. Sometimes misclassification due to trees or elevated terrain25Completeness and correctnessCompleteness: 286 missing buildings278 detected (97%)
- 51. 8 not detected (3%)Correctness: 316 buildings classified278 correctly classified (88%)
- 52. 38 misclassified (12%)26Alternative datasets?Address locations and building blocks missing in TOP10 (BRIDGIS)27Alternative datasets?BAG28Lessons learnedProcessing AHN2 is time consuming and requires large amounts of diskspace (4GB input generated 150GB output)
- 53. Many exceptions difficult to account for at forehand
- 54. AHN2 order process does not indicate percentage of NoData per subunit
- 55. AHN2 data has not been thoroughly checked (subunits entirely filled with NoData, wrong resolution, no crd sys)
- 56. TOP10NL is missing a lot of buildings (is not up-to-date)
- 57. Generalization of resolution is tricky (use block statistics, set resolution, snap to and extent correctly)29Questions?Xander BakkerGrontmij Nederland BV :: GIS & ICT - Team GIS :: http://www.Grontmij.nl :: +31 30 220 79 11http://software.grontmij.nl