In conjuction with the Applied Geomatics Research Group (AGRG)

1. Radiometric Calibration
of Digital Images
Capstone Project by
Sean Thibert
in conjunction with AGRG

2. Overview
• Project Goals
• Background Information
• Materials
• First Test (Linear)
• Python Script
• Second Test (Exponential)
• GNDVI Results
• Next Steps
• Conclusion

3.  Evaluate current methods for
radiometric calibration of a converted
“off-the-shelf” digital camera.
 Develop a time and cost effective
solution to integrate into AGRG’s UAV
data collection and processing
workflow.
Goals

4. Why is calibration necessary?
Satellite Sensors vs. Digital Cameras
Landsat Operational Land Imager (OLI) Nikon 1 J4 Digital Camera
• Highly specialized pieces of scientific
equipment
• Compensate for 705km worth of
atmosphere
• Have been operational for decades
• Refined laboratory-based
calibration methods
• Off-the-shelf camera modified to allow for
Near Infra-Red (NIR) light to be captured
• A relatively new mapping tool in
conjunction with Unmanned Aerial
Vehicles (UAV)
• Mission altitude < 90m, negligible
atmospheric effects
• Unique calibration requirements per
camera
• However, for all the same
reasons…

5. Why is calibration necessary?
Noise: Erroneous sensor measurements
Kelcey & Lucieer, 2012

6. Why is calibration necessary?
Vignetting: Pixel brightness changes radially
away from principal point as a factor of aperture
Kelcey & Lucieer, 2012

7. Why is calibration necessary?
Geometry: Lens Distortions
Kelcey & Lucieer, 2012

8. Why is calibration necessary?
Radiometric Calibration
◦ Conversion of recorded DN values (0-255) to
at-surface reflectance (%)

9. Why is calibration necessary?
Radiometric Calibration
◦ Compare datasets across sensors, days, and
locations with quantifiable units
DN x Calibration Coefficient = Reflectance (%)

10. Why is calibration necessary?
Radiometric Calibration
◦ Compare datasets across sensors, days, and
locations with quantifiable units
◦ Normalize images across one survey with
solar irradiance values
DN x Calibration Coefficient = Reflectance (%)
Reflectance (%) x Solar Irradiance (kW/m2)

11. Why is calibration necessary?
Radiometric Calibration
◦ Band-specific calibrations improve accuracy
of vegetation indices
◦ Focus of this project:

12. Materials
 Nikon 1 J4 converted digital camera
system for PrecisionHawk Lancaster
UAV

13. Materials
 Dataset of 29 images from Mosher’s
Corners, NS
◦ Acquired September 18, 2015
*all calibration images mimicked the above, except for altitu

14. Materials
 Ocean Optics Inc. JAZ portable
spectrometer

15. Materials
 Ocean Optics Inc. JAZ portable
spectrometer
Not this Jaz!

16. Materials
 Ocean Optics Inc. JAZ portable
spectrometer
• Measures spectral information from
350 – 1000 nm
• Adjusted for solar irradiance with a
Labsphere Spectralon Reference
Panel (95% Reflectance)
• SpectraSuite software for visualizing
and saving reflectance data

17. Materials
 Multicolour and Grayscale reference
targets
5% 20% 30% 40% 50% 60% 70%
80% 90%

18. Software
Fiji (Open source image
analysis package)
Open source distribution of
Python. Used GDAL, SciPy,
Numpy, and MatlibPlot
libraries
Photogrammetry
software used for
mosaicking
ArcMap 10.3 used for GNDVI
products

19. Where do they all add in?
1) Spectralon panel
reflects 95% of
incoming sunlight,
calibrating the
spectrometer

20. Where do they all add in?
2) Spectrometer
measures reflectance
of reference targets

21. Where do they all add in?
3) Camera captures
reference target in
several images while
in flight

22. Where do they all add in?
4) DN values plotted
against “true”
reflectance values to
determine relationship
Dependent variable
Independent variable

23. Linear Calibration Model
Ned Horning Public Lab post
◦ Straight-forward methods
◦ Inexpensive materials
 Used multicolour reference target instead
◦ Assumed linear relationship between DN and
reflectance

24. Results
Data for each camera band and target

25. Python Script
Input CSV with band-specific target reflectance
information and corresponding DN values
Script determines optimal regression equation for the data,
and stores it for use
User provides input and output location for images to be
calibrated
Script converts input image to 2D array, and applies
calibration equation to DN values for each band.
Writes new reflectance values to output .TIF image.

26. Results
y = 0.5032x - 37.773
R² = 0.7812
-20
0
20
40
60
80
100
0 50 100 150 200 250
Reflectance(%)
DN
NIR Linear (NIR)
Linear regression of NIR band
• Similar relationship with Green and Blue
• R2 = 0.7812

27. Results
y = 0.5032x - 37.773
R² = 0.7812
-20
0
20
40
60
80
100
0 50 100 150 200 250
Reflectance(%)
DN
NIR Linear (NIR)
X-intercept causing negative reflectance values

28. Results
y = 1.3469e0.0201x
R² = 0.9006
0
20
40
60
80
100
120
0 50 100 150 200 250
Reflectance(%)
DN
NIR Expon. (NIR)
Exponential regression is a better fit!
• Removes negative values
• R2 = 0.90057

29. Empirical Line Calibration
Model
 Methodology from Wang et al. (2015)
 “A simplified empirical line calibration method for sUAS-Based Remote
Sensing”, ASPRS
 Noticed similar exponential
relationship between DN and
reflectance
 Used grayscale reference target
instead of coloured
 More rigorous approach to data
collection

30. Empirical Line Calibration
Model
 Advocate calibrating each band
separately
 Performs a negative natural log
transformation on exponential
relationships to linearize the model
 Transforms the values back to
reflectance for calibration

31. Results
Data for each camera band and grayscale target

32. Results
Regression equations applied to each band from grayscale target panels
NIR Green
Blue

33. Results
Negative natural log-transformed data for NIR and Blue bands, due to their
use of an exponential regression curve

34. Final Calibration Equations
NIR:
Green:
Blue:
Applied to all images within the Mosher’s Corners dataset

35. GNDVI Results
1: GNDVI of 0.23 (indicative of vegetation)
2: GNDVI of 0.28 (indicative of vegetation)
1
2
1
2
1: GNDVI of -0.14
2: GNDVI of -0.03 (indicative of dirt)
Uncalibrated Calibrated

36. GNDVI Results
1: GNDVI of 0.07
2: GNDVI of 0.60 (indicative of healthy vegetation)
1
2
1
2
1: GNDVI of -0.061
2: GNDVI of 3.38 (indicative of shadows)
Note: Over and under exposed DN values causing shadows
Uncalibrated Calibrated

37. GNDVI Results
1: GNDVI of 0.27 (indicative of vegetation)
2: GNDVI of 0.23 (indicative of vegetation)
2
1
2
1
1: GNDVI of 0.04
2: GNDVI of -0.08 (indicative of dirt)

38. Next Steps
 Better calibration target
 Images recorded in RAW format
 In-situ testing
 Improve Python script:
◦ Batch process
◦ GUI
◦ Apply solar irradiance values

39. Conclusion
 Relationship between DN and
reflectance is exponential, not linear
 Empirical Line Method shows promise
 Most of the workflow can (and will) be
automated

40. References
• Berra, E., S. Gibson-Poole, A. MacArthur, R. Gaulton, A. Hamilton. “Estimation of the spectral sensitivity functions of un-modified
and modified commercial off-the-shelf digital cameras to enable their use as a multispectral imaging system for UAVs”. Remote
Sensing and Spatial Information Sciences, Volume XL-1/W4. Presented at the International Conference on Unmanned Aerial
Vehicles in Geomatics (2015)
• Haest, B., J. Biesemans, W. Horsten, J. Everaerts, N. Van Camp, J. Van Valckenborgh. “Radiometric Calibration of Digital
Photogrammetric Camera Image Data”. ASPRS 2009 Annual Conference, Baltimore, Maryland (2009)
• Horning, N. “Improved DIY NIR camera calibration”, PublicLab.org (2014). Accessed online at:
https://publiclab.org/notes/nedhorning/05-01-2014/improved-diy-nir-camera-calibration
• Kelcey, J. and A. Lucieer. “Sensor Correction of a 6-Band Multispectral Imaging Sensor for UAV Remote Sensing”. Remote
Sensing, Volume 4, Issue 5, pg 1462-1493 (2012)
• Laliberte, A., M. Goforth, C. Steele, A. Rango. “Multispectral Remote Sensing from Unmanned Aircraft: Image Processing
Workflows and Applications for Rangeland Environments”. Remote Sensing, Volume 3, pg 2529-2551 (2011)
• Lelong, C., P. Burger, G. Jubelin, B. Roux, S. Labbe, F. Baret. “Assessment of Unmanned Aerial Vehicles Imagery for Quantitative
Monitoring of Wheat Crop in Small Plots”. Sensors, Volume 8, Issue 5, pg 3557-3585 (2008)
• Ryan, R. and M. Pagnutti. “Enhanced Absolute and Relative Radiometric Calibration for Digital Aerial Cameras”. Photogrammetric
Week ’09, pg 81-90 (2009)
• Von Bueren, S. and I. Yule. “Multispectral Aerial Imaging of Pasture Quality and Biomass using Unmanned Aerial Vehicles (UAV)”.
New Zealand Centre for Precision Agriculture, Institue of Agriculture and Environment, Massey University (2013)
• Wang, C. & S. Myint. “A Simplified Empirical Line Method of Radiometric Calibration for Small Unmanned Aircraft Systems-Based
Remote Sensing”. Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 8(5) (2015).
• For more visual and technical information, please visit the following link from the Finnish Geodetic Institute:
http://www.kartverket.no/globalassets/kart/flyfoto/state-of-the-art-within-radiometric-correction-of-large-format-aerial-
photogrammetric-images.pdf