Airborne Camera Calibration for TerraPhoto

1
Airborne Camera Calibration for TerraPhoto

• Arttu Soininen
• Software developer
• Terrasolid Ltd

2
Why Airborne Calibration?

• Camera works in similar conditions as when
  collecting project data
• Airborne calibration is a true measure of how
  accurately the system works
• Derive parameters with TerraPhoto itself to
• Avoid hassle of converting values from a
  laboratory calibration document
• Derive values that work best for TerraPhoto

3
Calibration Result

• TerraPhoto camera calibration which works from
  one project to another
• We will still need to fine tune camera
  misalignment angles heading, roll and pitch for
  each flight session (IMU initialization offset)
• Rest of calibration stays constant

4
Calibration Site

• Site with easily identifiable features on the
  ground
• Paint markings on asphalt work nicely
• Site where most of the area is not covered by
  vegetation with significant height (trees,
  shrubs, tall grass)
• (Optional) Some control measurements on the
  ground to fix LIDAR ground model to be at the
  correct elevation
• Parking lots, airports, ...

5
Flight Pattern

• At least 4 flightlines at low altitude
• At least 2 flightlines at medium altitude (gt 2.0
  low)
• (Optional) 1 or 2 flightlines at high altitude (gt
  3.0 low)
• At least 20 images in total

Top view, low and medium flightlines
3d view, low and medium flightlines
6
Input Data for Calibration

• Laser model keypoints defining ground
• Trajectory (sbet.out, IPAS .pos, .lpo, .cte, ...)
• Image timing file
• Raw images
• Lever arm vector
• Timing offset (often 0.0)

7
Lever Arm Vector

• Defines a vector from input trajectory positions
  to camera focal point
• If input trajectory computed for GPS receiver,
  lever arm is from GPS receiver to camera
• If input trajectory computed for laser scanner,
  lever arm is from laser scanner to camera
• If input trajectory computed for camera, lever
  arm is 0.0, 0.0, 0.0
• Must be known beforehand
• 1 cm accuracy level
• Solving misalignment for each flight session
  compensates for small inaccuracy

8
Starting Values Calibration Document

• Use outside calibration document if TerraPhoto
  has a builtin conversion for it
• Leica RCD
• Rollei
• US / Applanix

9
Starting Values No Document

• If no calibration to convert, enter values for
• Image width
• Image height
• Plate width (image width)
• Plate height (image height)
• Timing offset (usually 0.0)
• Lever arm
• Orientation
• Principal point z
• Zero radius
• Leave all other fields as zero

10
Starting Values Principal Point Z

• Use pixels as the unit for principal point
• Usually the approximate focal length is known in
  millimeters and CCD pixel size is known
• Compute starting value as
• Value -Focal length (mm) / Pixel size (mm)
• Example
• Focal length is 50 mm
• Pixel size is 0.0068 mm
• Value -50 / 0.0068 -7353

11
Starting Values Zero radius

• Zero radius functions is the preferable model for
  radial lens distortion
• Choose a radius which fits inside the image
• Slightly less than half of the smaller dimension
  (width or height)
• Examples
• Image size is 7228 5428 pixels, use 2500
• Image size 3056 2032 pixels, use 1000

R
12
Calibration Workflow Phase 1

• Compare images and laser data in a camera view to
  make sure images fall roughly in the right
  location

13
Calibration Workflow Phase 2

• 1. Enter tie points, enter more tie points or fix
  bad tie points
• 2. Solve and apply misalignment angles using
  Output report command in Tie points window
• 3. Solve principal point z using Solve parameters
  command in Define Camera window and apply
• 4. Go back to step 1 until images have a good
  number of tie points (10 or more per image)

14
Calibration Workflow Phase 3

• 1. Solve and apply misalignment angles using
  Output report command in Tie points window
• 2. Solve all other solvable parameters using
  Solve parameters command in Define Camera window
  and apply
• 3. Go back to step 1 until there is no more
  improvement in the average mismatch distance

15
Timing Offset

• When flying is steady and straight, timing offset
  correlates with lever arm y component
• To verify timing offset, it may be beneficial to
  add one or two low altitude flight passes with
  unstable flying (changing roll all the time on
  purpose)
• TerraPhoto currently has no automatic solution
  for timing offset
• User has to try different values manually to find
  the one which gives the best results

16
Example Calibration Optech / Rollei

• Rollei 7228 5428 pixel camera
• Four flight passes at 600 m altitude
• 24 images, 8 cm pixel size
• Two flight passes at 1500 m altitude
• 6 images, 20 cm pixel size
• One flight pass at 2500 m altitude
• 3 images, 33 cm pixel size

17
Example - Passes Image Footprints
18
Example Site
19
Example Results in Tie Points

• Mixture of 8 cm, 20 cm and 33 cm pixel size
  images average tie point pixel size about 20 cm
• Starting average mismatch 9.07 cm
• Adjust positions statististics
• Final average mismatch 7.05 cm
• Avg heading change magnitude 0.005 deg
• Avg roll change magnitude 0.002 deg
• Avg pitch change magnitude 0.003 deg
• Avg elevation change magnitude 5.5 cm