Factory7

GantryB

GantryC

GantryF

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. (ADDENDUM)

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:
   • INITIAL_BODY_STATE = [ quaternion ; position ; quaternion temporal derivative ; position temporal derivative ]
   • IMU_TYPE = might be a manufacturer's part number or might be something else
   • GEODETIC_LATITUDE = Earth geodetic latitude in radians
   • CAMERA_OFFSET = position of the camera focus relative to the IMU's inertial axes in meters
   • T_cam = camera time stamps in seconds
   • CAMERA_TYPE = for camera calibration data, contact David Diel <ddiel at mit dot edu>