|
|
CMSC 828Z:: | |
Class Description :: Syllabus :: Reading List :: Projects :: Helpful Links :: | |
 |
|
CMSC828Z: Description of Project 1 - Radiometric and Color Calibration of the Cameras
|
||||
|
The goal of this project is to do a radiometric and color calibration of the 33 cameras that observed the scene. What does this mean? Although each camera is observing the same object in the scene, the intensity values that are recorded by the ccd of each camera and output in an image are different. The radiometric and color calibration tries now to find a consistent transform of the images recorded by each camera such that if the same object with diffuse surface reflectance is visible in two (or more) images, then the image intensity and color values at the corresponding locations should match, too. To be able to determine this transform, we observed a color checker chart as shown here |
||||
| ||||
| MacBeth Color Checker Chart | ||||
|---|---|---|---|---|
| This chart was moved all through the working volume and was shown to the different cameras at different angles and distances (see the following examples - the high-res image can be accessed by clicking on the low-res image): | ||||
| ||||
| As will be explained in the lecture about
different lenses, the light intensity is not the same across the lens
due to vignetting and the increasing angle between the principal light
ray direction and the ccd plane, so the calibration also needs to
account for this effect. Since we know the ground truth values for the
colors and intensities of the checker board (as given in the order
left to right, top to bottom by this MATLAB script which
returns the cie xyY values for each square as given by the
manufacturer and rgb values as measured from a scan of the color
chart), we can use these values to find the model parameters for the mapping between light rays intensities and pixel values. To convert between the different color spaces you can use the
following color conversion
tool. If you want to know more about these color spaces, take a
look at
Adobes webpage about color spaces. Note, since two thirds of the
cameras are gray scale, the calibration has to ensure that the
calibration is consistent in intensity space as well as color
space. The color cameras use the so called Bayer pattern (
see the linked paper for a more advanced algorithm to demosaic the
image). Part of the project will also be to calibrate the filter
response of the Bayer filter. One approach would be to use techniques from projective geometry and the supplied camera calibration information to find the position of the plane in the world for each frame of the sequence and then make sure that the colors of the chart have the same intensity in all the cameras that can see the chart.
The end result of this project will consist of a set of model parameters that describe the mapping from recorded image intensities to "real world" intensity and chromaticity values, where the intensity values should be consistent across all cameras and the chromaticities should be consistent among all the color cameras. |
