Structure-from-Motion, Grand Canyon style

Structure-from-Motion, Grand Canyon style

Presented at the Pacific Coastal and Marine Science Center USGS Santa Cruz, CA, Dec 7 2016

D612e176cb18a5190a722dc0313cb583?s=128

Daniel Buscombe

December 07, 2016
Tweet

Transcript

  1. Large-scale SfM: Grand Canyon style Daniel Buscombe, U.S. Geological Survey

    & Northern Arizona University, Flagstaff, AZ Rebecca Rossi, Utah State University, Logan, UT Paul Grams, U.S. Geological Survey, Grand Canyon Monitoring and Research Center, Flagstaff, AZ RSCC Workshop, USGS Santa Cruz, 7 December 2016 RM145, October 2015 ~10M points
  2. Photo: Matt Kaplinski, NAU Photo: Matt Kaplinski, NAU RM123, October

    2015 ~6M points
  3. Poles (no UAVs Allowed) GPS is unreliable GCPs with total

    station No man-made objects Remote
  4. GC Environments I: Sparsely vegetated sandbars RM70, October 2014 ~10M

    points
  5. GC Environments II: Heavily vegetated sandbars RM50, September 2014 ~23M

    points
  6. GC Environments III: Unvegetated sandbars RM30, September 2015 ~15M points

  7. GC Environments IV: Debris fans RM166, October 2014 ~38M points

  8. Complicated terrain Steep slopes RM213, October 2015 ~2M points RM91,

    October 2015 ~19M points
  9. Fast-growing riparian vegetation, several canopies RM202, October 2015 ~25M points

  10. RM145, October 2015 ~10M points Opportunities I: Automated feature extraction

    (mechanistic segregation) RM119, October 2015 ~43M points “intertidal” zone eroding bluffs vegetated aeolian surfaces water talus flat sand cut bank boulders
  11. Opportunities II: What’s the grain size of this debris fan?

    RM166, October 2015 ~115M points www.digitalgrainsize.org
  12. RM119, October 2015 ~43M points Opportunities III: Automated registration (Surveying

    GCPS is time-consuming … use rocks!?)
  13. Challenges I: Automated point cloud cleaning Temporary clutter People Water

    and boats RM137, October 2015 ~8M points
  14. RM91, October 2014 ~8M points RM91, October 2015 ~19M points

    Challenges II: Shadows can make automated classification difficult
  15. RM56, October 2014 ~29M points Shadows continued

  16. Spatially distributed uncertainty

  17. PySESA • Toolbox for statistical analysis of point clouds •

    Command-line python routines (with computationally demanding bits compiled) • Generic in design and application For SfM: • Point cloud decimation • Exploratory data analysis • Point cloud cleaning • Point cloud classification http://dbuscombe-usgs.github.io/pysesa/ Buscombe (2016) Computers & Geosciences
  18. http://dbuscombe-usgs.github.io/pysesa/ 4 different types of point cloud detrending

  19. From each data window Up to 34 metrics (spatial and

    spectral) http://dbuscombe-usgs.github.io/pysesa/
  20. RM202, October 2015 ~25M points

  21. 10 cm decimation Elevation

  22. Point density

  23. Roughness (locally detrended std.dev)

  24. Spectral strength

  25. 10 cm decimation Red

  26. 10 cm decimation Green

  27. Kurtosis Red

  28. Integral lengthscale Blue

  29. Fractal dimension Green

  30. Example: RM213, October 2014

  31. Range in elevation

  32. Roughness (spatial and spectral)

  33. Skewness (also kurtosis)

  34. Zeroth spectral moment (area under spectrum)

  35. Example: RM194, October 2014

  36. Thresholded Roughness

  37. Example: RM202, October 2015

  38. False color (spectral intercept, integral lengthscale and roughness)

  39. PySESA - SfM • Works on multiple dependent variables (z,

    r, g, b) • “red roughness”, “green skewness”, etc • Converts colorspaces (rgb to hsv, ycc, lab, etc) • Building in tools for supervised classification (needs calibration, lots of different types of data)
  40. RM145, October 2015 ~10M points Daniel Buscombe, U.S. Geological Survey

    & Northern Arizona University, Flagstaff, AZ Rebecca Rossi, Utah State University, Logan, UT Paul Grams, U.S. Geological Survey, Grand Canyon Monitoring and Research Center, Flagstaff, AZ Thanks: Joe Wheaton (USU), Joe Hazel, Matt Kaplinski, Ryan Lima (NAU) RSCC Workshop, USGS Santa Cruz, 7 December 2016
  41. None
  42. None
  43. First spectral moment (weighting higher frequencies)

  44. Integral lengthscale (range of roughnesses present)

  45. Minimum elevation (0.1 m decimation) (also mean, max)

  46. Point density

  47. Poles (no UAVs allowed) GPS is unreliable GCPs with total

    station Photoscan Pro
  48. None
  49. None
  50. None