Photogrammetry for Archaeology and Documentation

Tim Trombley

Educational Technologist

Media Center for Art History

Art History and Archaeology

Media Center for Art History

• learn.columbia.edu
• Projects from 20yrs+
• Resources for students
• Databases of resources and external projects

What is Photogrammetry

• Structure from Motion
• Create 3D models from photographs
• More specfic meanings out there, but this is what most people mean when they say photogrammetry

Comparison with other documentation approaches

• Total Station
• High-end Laser Scanning
• Handheld Laser Scanning
• Structured Light Scanning (SLS)

• Cheap!
• Quick
• Relatively Easy
• Flexible
• Active software development increasing accuracy and detail

Photogrammetry

• Cannot see results until post-processing
• Accuracy concerns and human error
• Computing requirements - RAM, GPU compute, CPU

Pros

Cons

Workflow

• Imaging
• Image Processing
• Photogrammetry Software
• Deployment

Imaging

• Camera Settings
• Coverage
• Use-case examples for different objects/settings
• Hope for an overcast day

Camera Settings

• Lens choices
• Consistent focal length (prime lenses) - software relies on EXIF data and lens profiles
• Sharpest focus possible (tripod highly recommended)
• Consistent exposure levels (can be adjusted in post to a point)
• Lowest ISO possible

Imaging Coverage

• Software can only process what it can see.
  Omissions = interpolation
• No such thing as too many photos (provided you can process them)

Imaging Use Cases

• Small object - Dome coverage
• Large object - Coverage/Height
• Small exterior/Trench - Coverage
• Large exterior/Site - Coverage, speed
• Drone: N/A - permits almost always required

Imaging Use Cases

Small object: Macchiozzo Inscription

Imaging Use Cases

Large object - Stronach Togate Torso

Imaging Use Cases

Small exterior/Trench: Onchestos Trench

Imaging Use Cases

Large Exterior: Macchiozzo

Imaging Use Cases

Large Interior: N/A

Will Attempt in Future likely with Fisheye Lens

Image Processing

• Consistent Lighting levels/exposure
• Masking - in Lightroom/Photoshop or Photogrammetry software
• High Quality Jpeg export - can use tiffs, but they are overly heavy in most cases
• Recommend working locally, not on network drive for reasons of speed

Photogrammetry Software

• Agisoft - popular, great community, relatively cheap, actively being developed
• Workflow:
  • align photos
  • refine/optimize cameras
  • generate dense cloud > export?
  • mesh
  • refine mesh
  • texture
  • export
• Quality settings can be adjusted at each step
• Higher quality = longer processing time and greater computing resources required, higher chance of crashes

Align Photos

• Basic placement of cameras relative to each other
• Looks for points in common
• Creates basic point cloud
• Will likely need straightened
• Lower quality settings may not align all photos
• Difficult photos may require manual point registration
• Some aligned photos may be erroneous and detrimental to larger model

Refinement and Optimization

• Gradual selection to remove garbage data
  • Reprojection error, Reconstruction uncertainty, Image count,  Projection accuracy
  • All values on sliders
• Optimize camera alignment after each change. Tighter alignment of cameras, less error
• Rinse and repeat

Generate Dense Cloud

• Most computationally intense step - looks for points in common between each photo
• Quality you choose plays a big role: 105 photos at the highest quality on my high-end workstation - 50hrs
• May result in 100s of millions of points
• Will almost always require cleaning of fly-away points
• Can register scale bars at this point for real world dimensions
• Could export points from this step to deployment with Potree or others or import into other software packages

Generate Mesh

• Inherently Interpolative
• Quality choice determines analytical viability and interoperability
• Can be exported and reimported for decimation and optimization
• Almost always a middle-man step

Create Texture

• Number of images used for texture
• Resolution of images
• UVW Wrapping

Deployment

• Audience
• Purpose
• Intended Devices

Deployment - Points

• Potree
  • Open-source (new release 3/12)
  • Octree loading - mobile friendly (to a point)
  • measurable, take sections, etc.
• Sketchfab - a bit awkward with points
• Other commercial solutions tend to be very expensive

Deployment - Meshes

• Sketchfab - huge library/community, VR/AR, strong support for cultural heritage, we have secured pro accounts for all CU emails
• Other commercial hosts and libraries
• 3D PDF - small models best, but viable in may situations, includes good tools
• Import into other software for compositing or analysis
• more flexible than points