Image Capture by Video4Linux@- Why don't you start computer vision research ?

@Capturing image is the first step to construct a vision system. There are many ways to capture image, e.g., using video capture device for PCI bus, USB and IEEE1394 etc. I think video input is still standard, I show the overview of image capture using capture device and Video4Linux (V4L) here. You can download a sample program. Using the program, you can start video processing fairly easily. I borrow V4L that is created by sincere efforts of many people to make this page. I really appreciate it.

yThings you needz

  Things you need to start image capture using V4L are

PC with Linux

Capture device with Bt848 or Bt878 (Bt8x8 is a capture chip)

Video camera or player

yEquipment I usedz

   To choose equipment sometime makes us confuse. I used low cost capture device and video camera you may find in your home.

 PCFPentium850MHz,Memory512MBALinux: Vine Linux 2.1CR

 Capture deviceF IO-DATA GV-VCP2M/PCI

 Video cameraFSony HandyCam DCR-PC100

The performance of ‚o‚b influences computational speed, but older PC with less memory can be used. Above capture device may not be available out of Japan, but you can use any capture device with Bt848 or Bt878. You may find devices from the sites in the last of this page. The cost may 50 USD to 100 USD. Every video camera with video output can be used. If you don't have it you can use video or DVD player (like PlayStation 2), but you can process only recorded materials.

yInstallation of device driverz

   After you install capture device into PCI bus, device driver controlling the capture device should be installed. An example of common procedure is as follows:

 (1) Find driver modules: i2c.o, videodev.o, bttv.o. You may find them in /lib/modules/2.2.x/mics(2.2.x is a version of kernel). You can use command "find" from root directory to find them.

 (2) Confirm device file. If there is no device file, e.g., /dev/video, make it using "mknod".

 (3) Install modules by /sbin/insmod.

 (4) Confirm installation by "dmesg".

If you use IO-DATA GV-VCP2M/PCI, you should use the option of card=2 to install bttv.o: /sbin/insmod bttv.o card=2. The capture device is recognized as "Hauppauge old". If you don't use this option, second composite input cannot be used. If you are working on Linux with kernel 2.4.x, i2c.o may be divided into 3 modules like i2c-core.o, i2c-dev.o, i2c-algo-bit.o. It is convenient to add the command for installing the models in /etc/rc.d/rc.local for the case of rebooting.

yCapture testz

  You can test capture device using distributed software like xawtv. I prepare the following sample program to construct vision system.

 show_video_v4l.tar.gz

Uncompress the file in a directory you use. Change default device file, /dev/video, in capture_v4l.h for your system. Mesa and glut must be installed to link executable file. They may be included in the recent distribution like Red Hat, but you can download them from the sites in links blow. If you succeed, you can find show_video_v4l in the directory. Use the options -v0(composite1) -v1icomposite2j -v2iS inputj according to the input channel you use (this option may not be used to other capture devices. Default is S input). You may see 320x240 video window on X window (left figure).  Option -s enables color extractor based on simple thresholding. Extracted pixels are shown by green. As you can see in right figure, they almost correspond to skin of Japanese. Push a left button of mouse on the window to end the program.

 @You may understand the overview of programming using V4L, Mesa and glut by the contents of sites in links below and source code of the sample. The first step to your own vision system may be to replace the color extractor of the sample by processing you need.

yLinksz

   I appreciate the following sites.

   Video for Linux resources page

   Video4Linux Programming

   The Mesa 3D Graphics Library
   GLUT - OpenGL Utility Toolkit

   IEEE1394 on Linux