This code uses the internal calibration parameters that are shipped with each Kinect, as opposed to extrinsically determining the calibration (the method used by most checker-board image calibration methods). Offline means that the calibration parameters can be captured once when the Kinect is plugged in, and then used in the processing phase without the Kinect on data that has already been acquired. Calibration is defined as the act of mapping from the raw sensor digital numbers (DN) in (pixel, pixel, DN) coordinates to world (x,y,z) coordinates. Registration is defined as mapping the RGB pixels onto the depth data.

Importantly, this registration maps RGB to depth, whereas all other registration algorithms I know of map depth to RGB. Since the data of interest with the Kinect is the depth data, it does not make sense to perform any unnecessary error-increasing mathematical operations on the depth data.

The code can be downloaded from https://github.com/mankoff/libfreenect/tree/offline_register and built with the following commands, assuming all dependencies are already installed. The easiest way to verify your are able to build this code is to install libfreenect through an existing package manager system, then over-write that with this custom build.

git clone https://github.com/mankoff/libfreenect/
cd libfreenect
git checkout offline_register
mkdir build
cd build
cmake ..
make
make install

Optionally, skip the make install line and access the new program in build/bin/. The new program is called “kinect_register” and when run without any arguments it prints usage instructions:

$ kinect_register

Data Formats (units: mm):
file.x: 640x480 double of x values
file.y: 640x480 double of y values
file.z: 640x480 integer of z values
file.xyz: ASCII file of x y z r g b. RGB only if PPM argument provided
file.reg.ppm: PPM shifted so pixels align with file.{x,y,z} data


% MATLAB
data = imread('file.pgm');
data = swapbytes(data);


If using my kinect_register program then reading a PGM is not absolutely necessary, as that program converts a single frame of depth data to an ASCII x,y,z file format. However, a single frame is noisy, and I suggest using the above PGM reader snippets to load several (or several hundred) PGMs and average all the good values of each pixel. Then, write out a new (averaged) PGM to be used as input to kinect_register. The snippets above can be used as templates for the writer function. The averaging will produce fractional numbers, but they will need to be rounded to whole numbers before writing the PGM.

If you have code to load 16-bit big-endian PGM files in another language, or a PGM writer function, feel free to share.

The role of Pine Island Glacier ice shelf basal channels in deep-water upwelling, polynyas and ocean circulation in Pine Island Bay, Antarctica

@article{Mankoff:2012The-role,
	Title = {{The role of Pine Island Glacier ice shelf basal
                  channels in deep water upwelling, polynyas, and
                  ocean circulation in Pine Island Bay, Antarctica}},
	Author = {Kenneth D. Mankoff and Stanley S. Jacobs and
                  Slawek M. Tulaczyk and Sharon E. Stammerjohn},
	Journal = {Annals of Glaciology},
	Number = {60},
	Volume = {53},
	Year = {2012},
        DOI = {10.3189/2012AoG60A062}}

Pine Island Glacier and Pine Island Bay

Pine Island Bay in the southeast Amundsen Sea, Antarctica, on 16 Nov 2008. Upwelling, melt-laden outflow plumes emerge from beneath the adjacent Pine Island Glacier ice shelf (top center) and mix in the bay waters. Warm red colors show sea surface temperatures more than a degree warmer than the near-freezing dark blue color. Cyclonic circulation in the bay is framed by the ice shelf, land ice and sea ice, in gray-scale with the darker shades colder. Landsat Enhanced Thematic Mapper Plus image, thermal infrared (channel 6H), subset of scene #LE72331132008321EDC00.

@article{Mankoff:2012The-role,
	Title = {{The role of Pine Island Glacier ice shelf basal
                  channels in deep water upwelling, polynyas, and
                  ocean circulation in Pine Island Bay, Antarctica}},
	Author = {Kenneth D. Mankoff and Stanley S. Jacobs and
                  Slawek M. Tulaczyk and Sharon E. Stammerjohn},
	Journal = {Annals of Glaciology},
	Number = {60},
	Volume = {53},
	Year = {2012},
        DOI = {10.3189/2012AoG60A062}}

AGU Poster: `Kinects as sensors in earth science: glaciological, geomorphological, and hydrological applications`

@conference{Mankoff:2011Kinects,
  Author = {Kenneth D. Mankoff and Tess Alethea Russo and
            Benjamin Kenneth Norris and Saffia Hossainzadeh and
            Lucas H. Beem and Jacob I. Walter and
            Slawek M. Tulaczyk},
  Title = {{Kinects as sensors in earth science: glaciological,
            geomorphological, and hydrological applications}},
  Address = {San Francisco, CA},
  Booktitle = {American Geophysical Union, Fall Meeting},
  Month = {December 5 - 9,},
  Note = {Abstract \#C41D-0442.},
  Year = {2011}}

Kinect Go Kit: Top level

Kinect Go Kit Pelican Case

I travel with two Kinects, two power supplies, and two computers in case one gets damaged or destroyed in the field, although only one computer is in the kit.

The Netbook is a cheap $240 computer running Ubuntu and the libfreenect software stack at a minimum. I have also found it useful to have more advanced data collection software (ROS, RGBDemo, RGBDSLAM), and some analysis software (CloudCompare, points2grid, Viewpoints, etc.). This netbook works fine for raw data dumps from the libfreenect ‘record’ program. It can run the more computationally expensive scene stitching algorithms such as RGBDSLAM, but it takes about 10 seconds per stitch, while a more powerful laptop (but still a few years old) can do it at 0.5 to 1 Hz. Since ‘record’ collects about 1.5 GB of data per minute, it is good to have a lot of free space on the hard drive.

The plugs and cables are shown laid out below, and in addition, some velcro straps are stored in that compartment, used to attach the Kinect to the tripod arm.

The Pelican 1510 case supports two levels, and the lower level looks like this:

Kinect Go Kit: Bottom level

Battery #1 is a 12 V 5 Ah sealed led acid battery. It provides >5 hours of Kinect runtime, about equal to the runtime of the netbook.

Battery #2 is 8 AA batteries (12 V), and underneath is an 8 AA battery holder and a battery charger. If I need to turn the Kinect on for a short amount of time and want to travel lightly, these will do.

Kinect cable layouts

As shown above the cord to the Kinect can be cut and alligator clips or some other electrical termination can be attached. I often have wall power and have attached clips to the detached plug so I can use it as originally intended. However, when in the field, the clips can connect directly to the 12 V battery or the AA battery pack.

Kinect mounted on tripod

The Kinect Go Kit above is close to the minimum necessary for fieldwork. Things that I would like in it, but are not yet, include:

• Tape measure
• Liquid container (tupperwear) and opaque liquid (or additive) so that any scene can have a defined flat surface
• Sling for under tripod to hold battery, netbook, protecting equipment and keeping it off the ground
• Counter weight for tripod arm

Additional tools I have found handy to have with me in the field include, but are not limited to, the following:

• External hard disk for backups
• Zip-ties to complement the Velcro straps
• Multimeter
• Spare notebook, perhaps with a more powerful CPU, for scene stitching
• Rope or other ‘image noise’ for scene stitching with RGBSLAM when working in environments that have ‘self similar’ scenes (no good tie points)
• Mounting systems for long term deployment
• Trashbags for environmental protection

There are a variety software interfaces to the Kinect. One high-level tool that is easy to use (binaries provided, no need to compile source, supports ‘scene painting’) is RGB-Demo. It is a good tool to start with if you want to work with the Kinect.

However, most Kinect software and calibrations so far have been developed by the robotics and computer vision communities. I am grateful for the work they have done, but those communities have different data needs than earth scientists. For example, quadrotor obstacle avoidance (link (PDF), link) has distance measurement errors that appear to be on order cm, but it still works fine as the helicopter avoids obstacles by an amount larger than the error.

Earth scientists should aim for a better model of the world than the one currently provided by the Kinect and its primary users. I suggest recording and storing the raw digital numbers (DN) from the Kinect rather than higher-level auto-calibrated real-world coordinates. It will require more post-processing, but storing the DNs will allow the data to be re-processed as better calibrations are developed. In addition, the low level recorder operates at 30 Hz and the higher level point-cloud products currently do not record data at that rate.

The best supported low-level interface is the LibFreenect Fakenect record program. It dumps the uncalibrated RGB and depth images to a folder at 30 Hz until you kill the process. Uncalibrated means both that the depth data is in sensor units, and that the depth and RGB images are not aligned. You can easily convert the depth data to real world x,y,z coordinates using existing published algorithms (link, link, link, and many others exist on the web), but importantly the raw data is stored and can be used with better calibrations in the future.

After processing the raw ‘record’ data, you can work with the point cloud data or DEMs using a variety of standard software for pointclouds, LiDAR, etc. I have had great success with CloudCompare and Poinst2Grid, in addition to custom code in MATLAB, IDL, and Python. A good list of software is available at the NSF OpenTopography site.

To work with the depth data to we initially use the following algorithms found on the various sites dedicated to Kinect hacking. The data provided by these algorithms is sufficient for certain uses, and for testing algorithms and visualizations, while better calibrations are performed.

DN to distance (source):

k1 = 1.1863d
k2 = 2842.5d
k3 = 0.1236d
Z = k3 * tan( double( DN ) / k2 + k1 )

XYZ to world (source):

Xres = 640
Yres = 480
FovH = 1.0144686707507438 (rad)
FovV = 0.78980943449644714 (rad)
XtoZ = tan( FovH / 2 ) * 2
YtoZ = tan( FovV / 2 ) * 2
X = ( X_pixel / Xres – 0.5 ) * Z * XtoZ
Y = ( 0.5 – Y_pixel / Yres ) * Z * YtoZ

Question or comments? Post below…

Can I unplug my fridge at night without ruining my food?