• Non-planar calibration rigs: The toolbox works perfectly with non-planar calibration rigs. For illustration, check the third calibration example posted on the main documentation page
.


• Robustness of convergence: The convergence algorithm may take a few more iterations than shown in the first example. However, the estimation process is now extremely robust, and stable. I have not found any example on which the optimization fails. If you do, please send me the data!
  
  
• Convergence movie: After calibration, the user can visually check the convergence of the optimization procedure by running the script file check_convergence. This script generates the successive intermediate 3D configurations of the calibration rigs (extrinsic parameters) computed during the optimization process (the variable param_list contains the list of intermediate calibration parameters, intrinsic and extrinsic).
  
  
• Reading images: The process of reading the images from disk is now a lot more automatic. The user only has to enter the images basename and format. The program automatically finds (and counts) all the images with that basename in the selected format and loads them. The images should still be consistantly numbered using the numbering scheme described on that page, but all numbers do not have to be consecutive (by step 1) anymore. It is perfectly fine is there are missing images. In fact, new images can be later on added at the missing slots, for more accurate calibration. Try it on example 1 by calibrating with the small set of images sparsely picked from the large list of images: Image1.tif, Image5.tif, Image9.tif, Image12.tif, Image15.tif (the local directory should not contain any other image from the list, otherwise, they will be read as well).



• Adding new images: If some number indices were initially left unused, then the additional images can now use those unused slots. Of course, the new images may still be numbered with higher number indices than the current larger number index (like described in the first example). Only one constraint: do not add images before the current first image (i.e. for the new image, do not use a lower index number that the index used for the current first image). In other words, as you add images, keep the index number of the first image identical. If you do not follow that requirement, there might be confusion on the correspondence between extracted corners, and image indices, and you may have to restart calibration from scratch. At any time during calibration, if you want to graphically see the current list of images, run the matlab script file mosaic.



• No-prompt use: Setting the global variable dont_ask to one lets you use the main toolbox functions (optimization + graphical display) in script mode without being prompted. This is also quite useful for including calibration calls in your own matlab script files. To set the toolbox in normal mode again, set dont_ask to zero (or clear the variable). For illustration, check examples 2, 3 and 4.



• Exporting calibration data to other formats: There is now a tool that allows to export calibration data (image corner locations + 3D world coordinates) to other formats that are compatible with Willson, Heikkilä and Zhang calibration codes. This option is useful for comparison purposes. The name of the matlab function is export_calib_data.m, and can be launched by clicking on Export calib data in the main calibration window. For more information, go to the first calibration example page.



• Converting Calibration results from Reg Willson's format : There is now a matlab script file willson_read.m that converts the calibration parameters generated by Reg Willson's calibration code into our convention. Visit page describing the parameters for more information.



• Selecting the intrinsic variables to estimate: The user can now take some of the camera intrinsic parameters out of the optimization. For example, it is possible to reject the principal point from the optimization by setting the variable center_optim to zero. It is also possible to reject the distortion model by setting the variable est_dist to the binary vector [0;0;0;0;0]. Other options as possible for the distortion model. For example, setting only the first entry of est_dist to one while leaving the other three entrie to zero would restrict the optimization to be done using a one parameter distortion model (first order radial distortion model). In addtion, it is now possible to run the optimization over the skew coefficient alpha_c by setting the binary variable est_alpha to one. For example, visit the first and second calibration example pages, and the page describing the calibration parameters.



• Now estimatates the Uncertainties: Now, the main optimization function go_calib_optim.m returns the uncertainties attached to the estimated calibration parameters (intrinsic and extrinsic)
Look at the function go_calib_optim_iter.m for more information.


• Stereo calibration: To calibrate a stereo system you may use the stereo calibration script calib_stereo.m. This script is not documented here, but a lot of information can be retrieved by typing help calib_stereo in the main Matlab window. Many users have successfully used it.

Back to main calibration page