1. Image Processing Collections

This page documents how to make and use image collections in Pyro.

1.1. SIVAL

This information provided by Grant Braught (braught AT dickinson.edu).

1. Download and unpack the unprocessed SIVAL images from www.cs.wustl.edu/~sg/accio/SIVAL.html

2. Copy into the SIVAL directory the two scripts:

• makeFakeCamreaFiles.bash
• makePyroDevices.bash

3. Edit the makeFakeCameraFiles.bash script to set the desired image size.

• Change the SIZE variable to the desired size.
• The ppm files are much larger than the .jpg images provided in SIVAL so the script scales them using the ImageMagic convert program.
• At 256x256 the resulting collection is just over 216MB.

4. Run the makeFakeCameraFiles.bash script.

• This produces a directory named FakeCameras.
• The FakeCameras directory will have one sub-directory for each of the SIVAL image directories (i.e. for each item photographed, e.g. cokecan).
• Inside each sub-directory will be 10 sub-sub-directories, one for each scene in which the item was photographed.
• The converted ppm images appear within the sub-sub-directories.

5. Edit the makePyroDevices.bash file to set the path to the FakeCameras directory.

• Change the BASEPATH variable to hold the path to FakeCameras on the machine where the FakeCamreas will be being used.
• The default value works with the PyroLiveCD and FakeCameras stored on the first USB drive.

6. Run the makePyroDevices.bash script.

• This will generate a .py file in each of the sub-sub-directories that defines the FakeCamera device for the set of images in that directory.

7. Fire up Pyrobot and have fun!

1.2. Stereo Images

Pyro now comes with a StereoCamera device that can do depth perception based on a pixel-to-pixel algorithm. It asks for the index of two other cameras that have already been started in Pyro.

If you have real cameras, simply load them, and start the StereoCamera like so (this assumes two Video for Linux cameras):

self.robot.startDevice("V4LCamera0")
self.robot.startDevice("V4LCamera1")
self.robot.startDevice("Stereo")

or from the command line:

pyrobot -r Test -d V4LCamera0,V4LCamera1,Stereo

Select indexes 0 and 1 (the defaults).

If you would like to test this feature using FakeCameras, then you will need to construct a special set of images, and then split them using yet another camera device. The reason that you cannot just use two Fake Camera objects is that the pairs of stereo need to be synced.

To sync the pairs of images, create a set of PPM files with the left and right images side by side, as show in this image:

A sample set of images is provided in the TutorialCameraStereo device.

Next, we need to split the images from camera0 into the left and right components. Load the TwowayCamera device, and select index 0 (the first camera) to split. There is also a FourwayCamera if you have a quad video input feed.

Finally, load the Stereo device and select the last two cameras to combine into a stereo image.

You can load it all up at the command line:

pyrobot -r Test -d TutorialCameraStereo,TwowayCamera,Stereo

and use index 1 and 2 (the split views of 0). Here is a sample of the above running. Depth is shown as grayscale---the whiter the pixel the closer to the camera it is perceived:

The algorithm, base code, and test images are provided at vision.stanford.edu/~birch/p2p/

Up: PyroModuleComputerVision