Data Interpretation

The data sets presented here represent cohesive inertial-visual scenes. All scenes come with color images, inertial data, and a MATLAB workspace containing several useful variables. The synthetic scenes, rendered in POV-Ray, also include depth maps and ground truth for the body motion. A simulated scene may have several inertial runs associated with it, since Monte Carlo simulation differs each time. I am also willing to make new simulations for the purpose of comparison with other estimation methods. Please take a look at the relevant papers:

D. Diel, P. DeBitetto, S. Teller. "Epipolar Constraints for Vision-Aided Inertial Navigation."
IEEE Motion and Video Computing. January 2005, pp. 221-228.

David D. Diel. "Stochastic Constraints for Vision-Aided Inertial Navigation." MIT Masters Thesis, January 2005.

You will need MATLAB and the Image Processing Toolbox to make full use of this data, although the images and inertial output can be read by other means. If you plan to use MATLAB, then you should get these helper functions.

  1. The inertial data is in international metric units, and can be read with ReadIMUdat().
  2. Newer versions of MATLAB recognize the .png image format. Use imread() or the helper function ReadColor().
  3. The depth maps must be read in their native 16-bits-per-pixel format. The least significant bit is equivalent to one millimeter. Use the helper function ReadDepth() for automatic conversion to meters.
  4. The workspace can be read with the command load('workspace'). The most important variables are explained below:

This material is based upon work supported by The Air Force Research Laboratory (AFRL) under Contract No. F33615-98-C-1201. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of AFRL. Please direct feedback to David Diel <ddiel at mit dot edu>.