Caspa camera boards
The Caspa series of Camera Boards from Gumstix
The MT9V032 sensor, at the heart of the Caspa FS camera board and the Caspa VL camera board, is not yet supported in the kernel, so a little work is necessary to get the camera up and running.
OMAP35x processors have dedicated hardware for capturing and processing data from image sensors. The Caspa camera sensor outputs raw 10-bit Bayer images which is transferred to the Image Signal Processor (ISP) via a parallel interface. The ISP contains various sub-modules that can be exported as Video for Linux (V4L2) devices in /dev. For additional hardware information see the external links section.
Contents
Quickstart
Hardware
Insert the white ribbon cable contact-side-up into the Caspa connector and contact side down into the Overo connector.
Using a small flat-head screwdriver, loosen the lensholder set screw and rotate the lens to adjust the focus.
Install Pre-built Kernel
The easiest way to get started is to replace the kernel and modules of an existing image. This has been tested on an Overo Fire with a Tobi expansion board using the desktop image available here.
Get the kernel and modules.
$ wget http://cumulus.gumstix.org/images/angstrom/misc/caspapx/uImage-2.6.34 $ wget http://cumulus.gumstix.org/images/angstrom/misc/caspapx/modules-2.6.34.tgz
Copy to a bootable microSD card.
$ tar -xf modules-2.6.34.tgz -C /media/mmcblk0p2/ $ cp uImage-2.6.34 /media/mmcblk0p1/uImage $ cp uImage-2.6.34 /media/mmcblk0p2/boot/uImage
Alternatively, if you have a network connection, you can copy the files without removing the microSD.
$ tar -xf modules-2.6.34.tgz $ scp -r lib root@overo:/ $ scp uImage-2.6.34 root@overo:/boot/uImage $ scp uImage-2.6.34 root@overo:/media/mmcblk0p1/uImage $ ssh root@overo 'depmod; shutdown -r now'
Test
From the serial console:
# export DISPLAY=:0.0
# gst-launch v4l2src ! xvimagesink or # mplayer tv:// -tv driver=v4l2:device=/dev/video0
From the desktop:
<left-click> Applications > Multimedia > Cheese
If video0 doesn't appear in /dev you might have to load the driver manually.
# depmod # modprobe mt9v032
Bitbake
You can build the kernel and modules using bit-bake. This allows you to incorporate the kernel with camera driver into an image of your design.
$ cd ~/overo-oe/org.openembedded.dev $ git pull origin overo $ bitbake linux-omap3-caspapx
When the build completes, the kernel and modules will appear in ~/overo-oe/tmp/deploy/glibc/images/overo.
Customization
This section describes how to customize the camera driver and kernel for your needs.
2.6.34
Get the kernel source that is used in the Gumstix kernel recipe (~/overo-oe/org.openmebedded.dev/recipes/linux/linux-omap3_2.6.34.bb) and create a new branch from the recipe SRCREV.
$ cd ~ $ git clone git://www.sakoman.com/git/linux-omap-2.6.git $ cd linux-omap-2.6 $ git checkout -b 2.6.34 cb89736af28f583598e49a05249334a194d00f1d
Get the patch.
$ wget http://cumulus.gumstix.org/sources/mt9v032-2.6.34.patch $ patch -p1 < mt9v032-2.6.34.patch
Get the kernel configuration. This is the same one used for the linux-omap3-caspapx image.
$ wget http://cumulus.gumstix.org/sources/mt9v032-2.6.34.defconfig $ cp mt9v032-2.6.34.defconfig .config
If you haven't done so yet, you'll need to set up your build environment. Build the console image. This will ensure that the cross compiler, binutils, and libraries need to compile the kernel are installed.
$ bitbake omap3-console-image
Add the cross compiler to your path and configure the kernel. The cross compiler location depends on the architecture of your build machine.
32-bit processors
$ export PATH=/home/<username>/overo-oe/tmp/sysroots/i686-linux/usr/armv7a/bin:${PATH}
64-bit processors
$ export PATH=/home/<username>/overo-oe/tmp/sysroots/x86_64-linux/usr/armv7a/bin:${PATH}
Build kernel and modules
$ make ARCH=arm CROSS_COMPILE=arm-angstrom-linux-gnueabi- uImage $ make ARCH=arm CROSS_COMPILE=arm-angstrom-linux-gnueabi- modules
Install modules
$ mkdir ~/linux-omap-2.6/modules $ make ARCH=arm CROSS_COMPILE=arm-angstrom-linux-gnueabi- INSTALL_MOD_PATH=./modules modules_install
Other Kernel Versions
The ISP code used for the 2.6.34 patch came from the arago-project. This is the OMAP3 PSP kernel integration/staging tree
$ git remote add git://arago-project.org/git/projects/linux-omap3.git aragoOMAP3 PSP kernel integration/staging tree $ git fetch arago $ git checkout -b arago
The V4L2 subdevice framework offers a way to configure the Image Signal Processor pipeline from user-space. This is currently under active development and users wishing to make use of this should visit the Media controller development repository here. This contains the bleeding edge Video for Linux media framework, ISP, and MT9V032 drivers.
MT9V032 Driver
If you want to take advantage of an unimplemented sensor feature, change defaults, etc., you need to modify the driver source. After you've made a change you can use a script like the following to quickly test your new driver.
#! /bin/bash
TARGET_OVERO="root@10.0.1.15" make -j8 ARCH=arm CROSS_COMPILE=arm-angstrom-linux-gnueabi- modules ssh $TARGET_OVERO 'rmmod mt9v032' scp drivers/media/video/mt9v032.ko $TARGET_OVERO:/lib/modules/2.6.34/kernel/drivers/media/video/ ssh $TARGET_OVERO 'insmod /lib/modules/2.6.34/kernel/drivers/media/video/mt9v032.ko' ssh $TARGET_OVERO 'mplayer -display :0.0 tv:// -tv driver=v4l2:device=/dev/video0'
Image Signal Processor
There is some code that you can't modularize and modifying means rebuilding the kernel. Such is the case for the ISP code which is used by board-overo.c for hardware configuration. Another script like that in the previous section can be very useful.
This script rebuilds the kernel, copies it to the COM, and resets:
#! /bin/bash
TARGET_OVERO="root@10.0.1.15" make -j8 ARCH=arm CROSS_COMPILE=arm-angstrom-linux-gnueabi- uImage; scp arch/arm/boot/uImage $TARGET_OVERO:/boot/uImage; scp arch/arm/boot/uImage $TARGET_OVERO:/media/mmcblk0p1/uImage ssh $TARGET_OVERO 'shutdown -r now'
This is useful for adjusting the filters, white balance, and color blending coefficients used by the ISP preview module. The preview module is used for doing hardware conversion from 10-bit Bayer to YUV. If you are having color problems - especially under certain lighting conditions but not others - look at isppreview.c and the gamma correction tables in drivers/media/video/isp/
Tuning
Module Parameters
By default, auto exposure, auto gain, and high dynamic range are all enabled. You can disable any or all of these to suit your needs by reloading the driver.
Get a list of driver parameters.
# modinfo mt9v032 filename: /lib/modules/2.6.34/kernel/drivers/media/video/mt9v032.ko license: GPL author: Ignacio Garcia Perez <iggarpe@gmail.com> description: mt9v032 camera sensor driver srcversion: A46920FD64C35A2645E0D54 alias: i2c:mt9v032 depends: vermagic: 2.6.34 mod_unload modversions ARMv7 parm: sensor_type:Sensor type: "color" or "mono" (charp) parm: auto_exp:Initial state of automatic exposure (int) parm: auto_gain:Initial state of automatic gain (int) parm: hdr:High dynamic range (int) parm: low_light:Enable companding (int) parm: hflip:Horizontal flip (int) parm: vflip:Vertical flip (int)
Depending on the light level, auto exposure might reduce the framerate as low as 15 frames per second. You can disable auto exposure like so:
# rmmod mt9v032; # insmod /lib/modules/2.6.34/kernel/drivers/media/video/mt9v032.ko auto_exp=0
If this makes your image too dark, you can disable auto gain and enable companding mode (more information here).
# rmmod mt9v032 # insmod /lib/modules/2.6.34/kernel/drivers/media/video/mt9v032.ko auto_exp=0 auto_gain=0 low_light=1
