Roadmap

• Improve georeferencing / coordinate system throughout entire pipeline
  • Needs to be an integrated part of the the pipeline, not a single process at the end
  • Feature branch: ****
• create benchmarking

Merge existing and proposed changes into the cmake fork

• add any OO improvements from other work
• remove junk code
• update argparse arguments
• Orthophoto .tif generation
• update Dockerfile (Link? ****)
• fix GCP reading order
• add GDAL specifications for GCP lists
• add XYZ point cloud output
• use mkdir_p() everywhere
• optimize prepare_objects (Doesn't exist)
• improve logging and messages
  • Add benchmarking and generate a log file
• rerun_all functionality
• remove depreciated code (ccd_defs, etc)
• other clerical changes (README, etc)
• replace slow code (i.e. PMVS) (Feature branch: ****)
• Implement testing framework. See https://github.com/dakotabenjamin/OpenDroneMap/tree/nose_tests

Push cmake fork to gh-pages

• implement testing protocol
• update licenses
• release odm v0.1

Texturing/blending (contracted)

• Add image blending and exposure adjustments (Feature branch: ****)
• Improve image stitching

Video/SLAM reconstruction (contracted)

• ORB vs. LSD SLAM approaches (Feature branch ****)

Produce a benchmarking file

• See this issue
• Pull Request

Update dependencies

• pyexiv2 is depreciated in favor of gexiv2

Long Term Improvements:

Processing pipeline improvements:

• Semi-global matching for denser point clouds (http://cdcseacave.github.io/openMVS/ ?) (Feature branch: ****)
• Replacement of lastools with PDAL
• Add optional pre-calibration tool, maybe with editable library of lens parameters derived from checkerboard calibration process
  • Formalization of Pau, Dakota, and other’s work
  • Formation of dedicated library, perhaps borrowing from existing database

Processing pipeline augmentation:

• Point cloud partitioning / classification
• Improvement of meshing (treat buildings differently than vegetation than ground, etc.)
• DEM generation, see and integrate:
  • https://smathermather.com/2016/10/21/filtering-buildings-and-vegetation-from-drone-derived-surface-models/
  • https://smathermather.com/2016/10/15/finding-peace-finding-the-grounddrone-flights-for-hydrologic-modeling/

Web improvements

• Development of Python node API — node-OpenDroneMap
• Development of web front end(s) — WebODM
• Integration with OpenAerialMap

Messy additions from SOTM 2016

Feature Requests:

Core

• Point clouds

  • Semi-global Matching
    • https://github.com/mapillary/OpenSfM/tree/dense
    • Status:
      • Due October 1st
  • Point cloud classification
    • Progressive morphological filter via PDAL
      • http://www.pdal.io/stages/filters.ground.html
      • Status:
        • No current work, but simple addition of PDAL functionality. Need build process which allows for necessary dependencies
      • Advantage: Allows for classification of ground point cloud for partitioning of mesh creation / quality, hydrological applications, etc.
      • Classifications based on machine learning of input raw images

• Mesh

  • Change meshing from Poisson to alternate mesh (not water tight?)
  • Use different mesh approaches based on classification of data
  • Identification of flat surfaces
    • e.g. http://blog.spotscale.com/2016/03/29/make-it-flat/
  • 3D / 2D Breaklines
    • Specialized interface or iD
    • OSM or custom
    • Possibly integrate with OSM workflow and iterate orthos
    • Alternatively digitize directly on raw images

• GPS positioning input alternatives

• Atmospheric correction

Texturing

• Add vignetting correction to toolchain
• BRDF?
• Fix interpolation gradients (CIR use case)

Output

• Crop tiff output to high probability good data (convex or concave hull) by default (flag to output all)

Optical distortion

• Correct for known cameras
  • Fixes for wide angle
  • Optimization of processing when using corrected imagery

API

Web

Integration

Error calculation / reporting

Logs

• Progress Status
  • Markers in log for process for rubber sheeting to state
  • Image Matching refinement of progress

Optimization

• Parallelization of

  • resize
  • focal length

• GPU

• Incremental SfM

• Clusterability

  • Possibilities:
    • Shared file system + celery workers
    • Smarter approach using data localization
    • PySpark
      • Working in terms of data frames
      • Lambda expressions that can be sent out to separate machines
      • Process steps need to be more granular
    • Serializable intermediate representation

• Find automated rubber sheeting solution for rubber sheeting to other datasets

Future:

Use Cases

• Manned aircraft app
• iPhone processing
• Single tiled aerial of OAM outputs
• Fly around house and create model
• Streetview

POSM / OAM-PacDID / ODM

• How much of the OAM tiling has been solved by POSM (Cristiano)?

  • Internal tiling with VRTs in original projection with pyramids
  • Seth advocates running OAM on lambda and paying usage
  • Need to retile and pyramid the existing imagery
  • Re organize to allow for mosaic
  • If don't want mosaic, recreate function for retiling and render

• PacDID needs

  • WMTS (ArcGIS/QGIS)

  • AOI

    • Coordination of data collection
    • Coordination of existing data (index current imagery)
    • Need for Flight planning info

  • Next phase -- how can it help projects (Nate)

• What is POSM / ODM integation stuff

  • Question of nodeODM / WebODM

• OAM Opportunities

  • Make easy to deploy
  • Documentation