| We've got more cameras to
calibrate for the year. This time "feature based" methods were used. Namely,
we used a theodolite to accurately measure landmark positions. The landmark
position in the image was identified and a world-image point correspondence
was established. A Tsai-like algorithm was used to simultaneously calibrate
internal/external parameters. The parameters were found by global optimization
via Levenberg-Marquardt method.
One fixed black/white camera on the Wean Hall could see no landmarks at all (the laser beam of the theodolite was occluded). To calibrate it, a calibrated pan/tilt camera was set to have roughly the same field of view as the static b/w camera. Image features were identified in both images and correspondence were established. With the aid of the pan/tilt camera, the b/w camera was satisfactorily calibrated. To calibrate an infrared camera, a coke can containing hot water was placed on the feature point. The hot coke can gave a distinctive feature point in the scene. Here is the code for the calibration and an example input file. If you are calibrating a camera mounted on top of a pan/tilt head you may consider using this program. Built upon Tsai method, the program added the calibration of a mounting matrix and consideration of the pan/tilt angles. Friends who used it said good things about it and came back asking for it. There are a set of switches and initialization needed to be done at the beginning of the input file. After the settings you put each pair of 3D-2D corresponding points in the format of [X Y Z u v] in a row. Sorry no more documentation about the code. And I only tested it on Windows. The code evolved from an implementation of Tsai's method by R. Willson. |