Showing posts with label ubuntu. Show all posts
Showing posts with label ubuntu. Show all posts

Tuesday, February 26, 2013

Raspberry PI, Raspbian, XBMC and eGalax 7 inch touchscreen

Hello!

I have spent some time lately trying to find a solution to get my 7 inch eGalax touchscreen to work with  Raspbian(Debian Wheezy) in XBMC 12 Frodo and finally got it working as I wanted.

My Setup
  • Raspberry PI model B: ~30$
  • 7 inch display with touchscreen for car rear view camera, from eBay(touchscreen is connected to one USB port): 80$
  • HDMI male to HDMI male connector(from eBay): <2$
  • 4GB SDHC class 4 card
  • 12V(500mA) AC to DC adapter for powering the display
  • 5V(1A) microUSB AC to DC converter for powering the PI
  • USB keyboard


Edit:  Download the latest image from the top right corner of this blog(username: pi, password: a).

Here is what you need to do in order to have a system with Raspberry PI, Raspbian OS and XBMC 12 Frodo stable with eGalax touchscreen working correctly(which means axes calibrated and click working with just one tap&release action):


1. Get latest Raspbian image from here and flash it to an SD card.

2. Build your own kernel with eGalax touchscreen support, like in this post(you will only need to replace kernel.img file and /lib/modules and /lib/firmware folders on the SD card).

3. Build XBMC 12 on Raspberry PI using this tutorial.
Note: After downloading XBMC archive, get this archive and unpack it anywhere.
Apply patches to xbmc files:
cd <patches_folder>
patch -p1 <path_to_xbmc>/xbmc/input/linux/LinuxInputDevices.cpp < LinuxInputDevices_cpp.patch
patch -p1 <path_to_xbmc>/xbmc/input/MouseStat.cpp < MouseStat_cpp.patch
patch -p1 <path_to_xbmc>/xbmc/input/MouseStat.h < MouseStat_h.patch
4. Touchscreen calibration.
Create a new file /home/pi/touchscreen_axes_calib on Raspberry PI. It will contain four values for the axes calibration and one value for swapping axes.
The simplest way to swap axes is to switch the four wires cable plug's orientation which comes from the touchscreen to the touch controller.

Here is how the calibration was done.

the original behavior(no calibration)

In the picture above, we see that "touch panel values frame" differs from "touch panel physical size frame". When we are pressing the touch we are moving in the "touch panel physical size frame" but when the touch screen is not calibrated the arrow from XBMC is in another place.
  • "touch panel physical size frame" is the screen starting from (0,0) on the left top corner and going to (width, height) in the right bottom corner.
  • "touch panel values frame" is the frame which contains all the number the touch controller is giving.
We see that these frames differs a lot. Our main scope is to overlap the "touch panel values frame" to the "touch panel physical size frame".

In order to do this we need to do three steps(the third one is done in software):
a. Scale the value read from the touch driver x and y) in order to fit 0->width range and respectively 0->height range of the "touch panel physical size frame" the scale value for x axis is:
                       "touch panel physical size frame" width
calib_x_fact = -------------------------------------------------
                            "touch panel values frame" width


                       "touch panel physical size frame" height
calib_y_fact = -------------------------------------------------
                            "touch panel values frame" height

"touch panel values frame" width and height are coming from your XBMC resolution(I have width=1280 and height=720).
"touch panel physical size frame" width and height are a little more trickier to find but nothing hard. In step 2 above, you have calibrated the touchscreen in XFCE. You got some values returned by xinput_calibrator, something like:
Section "InputClass"
    Identifier   "calibration"
    MatchProduct    "eGalax Inc. USB TouchController"
    Option    "Calibration"    "1977 32 1893 131"
EndSection
In my case,
"touch panel physical size frame" width is 1977 - 32 = 1945
"touch panel physical size frame" height is 1893 - 131 = 1762
Now, compute the values and put them in /home/pi/touchscreen_axes_calib file

b. Translate the "touch panel values frame" to the left and up, to match "touch panel physical size frame".
I didn't find a logical method to do this, because we don't know exactly "where is" the "touch panel values frame", so, in order to find calib_x_d and calib_y_d you have to first set them both to zero and then start XBMC. Now, put some sharp pointer on the screen and observe the distances between the cursor on the screen and your pointer's position. Try to approximate these x and y deviations(measured in pixels) and put them in the /home/pi/touchscreen_axes_calib file.

c. Revert direction of axes. This is done in the software(from patches).

5. Math behind.
To accomplish these transformations the following formula was implemented in the file
xbmc/input/linux/LinuxInputDevices.cpp
pointer.x = value_read.x * calib_x_fact + calib_x_d;
pointer.y = value_read.y * calib_y_fact + calib_y_d;

After I have successfully calibrated the touchscreen I have discovered that single click was not possible from the touchscreen, just double click. After digging through the code, I have found that this was caused by drag action which was triggered because the previous values of the touch were far(more than 5 pixels) from a new press. For example, at the start of the program, cursor is set at 0,0 coordinates; if user is trying to press a button, let's say at 100, 300, the program(XBMC) will calculate the distance between these two points and will find out that this is greater than 5.
Pythagorean theory:
(100-0)x(100-0) + (300 - 0)x(300-0) is greater than 5x5 XBMC will treat this as a drag event.
This drag issue is not caused when you double click, because the previous point in the second click action is very close to the second click point. This also works for mouses, because the previous value of the pointer is always very close to the new value of the pointer(because mouse's pointer drags on the screen and it doesn't jump - so each new value is very close to the previous one).

I have developed an algorithm to avoid this issue:
When the user is pressing the screen(x,y), the touch values are being set to (screen_width+1, screen_height+1 -> outside of the visible screen) just at the first event read(which is BTN_TOUCH PRESS).
After this event, the program will receive multiple X and Y absolute values events. The first two events, one for X and one for Y are used to set the previous X value, respectively previous Y value to the current X respective current Y values. And from now on distance is measured and this is preventing no unwanted drag action.
The user's finger/pointer will not stay at a single point, because the touchscreen's lack of precision, so it will move around 5-6 pixels in x and y directions.
I have also set the click distance to 7. You can change this by changing click_confines value in xbmc/input/MouseStat.h. Originally it was set to 5, but this is not very good for touch screens(I had to click with a sharp pointer and with my nail always, but with a value of 7 I can click with my finger with a slight touch -> really nice).

Enjoy!

Saturday, January 19, 2013

Adding 7inch display with touchscreen to Raspberry PI

Hi!

First thing I got in mind when seeing Raspberry PI was "car PC project".
The targeted display was 7 inch with touchscreen. I have found a lot of displays on Ebay.

I have got myself one for 85 dollars with free shipping(this; if it is not available any more you can search "reversing driver board hdmi" on ebay and you will find others). The display driver board has hdmi input and an on board resistive touchpanel with usb controller board.

It took less than a month to receive it(in Romania). After unpack, it worked out of the box with Ubuntu 12.10(display + touchpanel) and with Windows, but for Windows I had to install some drivers also received in the package.

I have installed latest Raspbian image on a SD_Card and tried it on my Raspberry PI model B, but the touchpanel didn't show any input. After searching a lot I have decided that I have to recompile the raspbian kernel and add support for touchpanel. This sounded very new to me but it seemed to be an easy task.

First thing, I have run lsusb to see the touch controller type(on RaspberryPI):
pi@raspberrypi ~ $ lsusb
Bus 001 Device 002: ID 0424:9512 Standard Microsystems Corp.
Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
Bus 001 Device 003: ID 0424:ec00 Standard Microsystems Corp.
Bus 001 Device 004: ID 1c4f:0002 SiGma Micro Keyboard TRACER Gamma Ivory
Bus 001 Device 005: ID 0eef:0001 D-WAV Scientific Co., Ltd eGalax TouchScreen
Last device is the touch controller, from eGalax.

Edit: If you don't want to build the kernel by yourself, you can download mine from here. After his, you have to replace file /boot/kernel.img and /lib/firmware and /lib/modules/ on the SD card.

Building a new kernel(in UBUNTU 12.10).
Get kernel sources.
wget https://github.com/raspberrypi/linux/archive/rpi-3.6.y.tar.gz
tar -zxvf  rpi-3.6.y.tar.gz
Install some dependencies.
sudo apt-get install git libncurses5 libncurses5-dev qt4-dev-tools build-essential
Install toolchain.
The best way to do the kernel compilation is on a Desktop/Laptop machine, which will be much more fast than on the Raspberry PI. I have did this in Ubuntu 12.10:
sudo apt-get install gcc-arm-linux-gnueabi
After download of the kernel archive has finished unpack it and then navigate with terminal to the extracted folder.
Be sure thaat the sources objects are cleaned. Type:
make mrproper
Create a folder for the generated kernel:
mkdir ../kernel
Generate the .config file:
make O=../kernel/ ARCH=arm CROSS_COMPILE=/usr/bin/arm-linux-gnueabi- bcmrpi_cutdown_defconfig
Configure the kernel:
make O=../kernel/ ARCH=arm CROSS_COMPILE=/usr/bin/arm-linux-gnueabi- xconfig
In the opened window press the | button to collapse all items. Then, navigate to Device Drivers->Input Device Support->TouchScreens and select it. Here, be sure to check also your touch screen controller if it is other than eGalax, or if it is not selected. Now press save.

With the changes being made you can now compile the kernel:
make O=../kernel/ ARCH=arm CROSS_COMPILE=/usr/bin/arm-linux-gnueabi- -k -j3
Note: -j3 option from the end means to enable parallel build. The number should be number of cpu cores + 1(I have dual core cpu).

The build took about 20 minutes on my PC. After the build completes, you will have the new kernel in ../kernel folder, created above.

Create the kernel image:
cd ../
git clone git://github.com/raspberrypi/tools.git
Note: You need to have git installed.

Navigate to tools/mkimage and then run:
./imagetool-uncompressed.py ../../kernel/arch/arm/boot/Image
This command will generate the kernel image(kernel.img file).

Build modules:
Go back to the ../kernel/ folder.
mkdir ../modules/
make modules_install ARCH=arm CROSS_COMPILE=/usr/bin/arm-linux-gnueabi- INSTALL_MOD_PATH=../modules/

Replace the kernel:
Get latest firmware:
wget https://github.com/raspberrypi/firmware/archive/next.tar.gz
tar -zxvf next.tar.gz
In the small partition(/boot) do:
  • replace /boot/bootcode.bin with firmware-next/boot/bootcode.bin
  • replace /boot/kernel.img with the previously created kernel image
  • replace /boot/start.elf with firmware-next/boot/start.elf
In the big partition(/root) do:
  • replace /lib/firmware with <modules_builded_above_folder>/lib/firmware
  • replace /lib/modules with <modules_builded_above_folder>/lib/modules
  • replace /opt/vc with firmware-next/hardfp/opt/vc/
Now your card should contain the new image. Safely eject your SD card and then unplug it from the card reader and then put the card in Raspberry PI and start X(startx). Plug the touch controller in one usb and check if you can move the cursor(or you can start with he touch already plugged in).

After I have started X, my touch input was working but the axes were switched and also not calibrated.

Calibration for the touchscreen(in Raspberry PI). 
Note: The next steps are performed in the Raspberry PI's Debian Wheezy. This is a method for calibrating the touchscreen which will work just for Xserver and Xserver based applications.

Install xinput_calibrator.
Install some dependencies:
sudo apt-get install libx11-dev libxext-dev libxi-dev x11proto-input-dev
Download xinput_calibrator somewhere in the Raspberry PI's folder structure.
wget http://github.com/downloads/tias/xinput_calibrator/xinput_calibrator-0.7.5.tar.gz
Unpack it and then navigate to the unpacked folder and then install it using:
./configure
make
sudo make install
After this step you should run xinput_calibrator(from Xserver terminal console: first startx then open console and then run it).
xinput_calibrator
Follow the on screen instructions(touching some points on screen) and after calibration is complete you will receive a message like this:
Calibrating EVDEV driver for "eGalax Inc. USB TouchController" id=8
    current calibration values (from XInput): min_x=1938, max_x=114 and min_y=1745, max_y=341

Doing dynamic recalibration:
    Setting new calibration data: 121, 1917, 317, 1741


--> Making the calibration permanent <--
  copy the snippet below into '/etc/X11/xorg.conf.d/99-calibration.conf'
Section "InputClass"
    Identifier    "calibration"
    MatchProduct    "eGalax Inc. USB TouchController"
    Option    "Calibration"    "121 1917 317 1741"
    Option    "SwapAxes"    "1"
EndSection

For Raspbian you have to create a file:
sudo nano /usr/share/X11/xorg.conf.d/01-input.conf
Add in this file the content above(starting with Section "InputClass" line) and then save it(ctrl+O).

Note:
Please make sure that you don't have sections like
MatchProduct    "eGalax Inc. USB TouchController"
in other files from /usr/share/X11/xorg.conf.d/ folder(highest number files are processed last, thanks to Jasmin).

Now touchscreen should be calibrated and after reboot it will keep the settings.
Once, I had to run xinput_calibration again in order to have the pointer to the desired points. You can update the numbers given by the xinput_calibration utility in the
usr/share/X11/xorg.conf.d/01-input.conf file in order to have the best calibration at boot.

Soon I will add some pictures.

Andrei

Wednesday, November 28, 2012

STM32F3-Discovery Usart with printf

Hi,

The next step after setting up the Development Environment for STM32F3-Discovery was to communicate with computer via serial port.
I have added usart support to my board using USART2 module, which had TX connected to PA2 pin and RX to PA3 pin(both with alternate function set to 7).

The next step was to connect the pins to PC serial port. I have used a MAX3232 module(note that you need a chip with 3.3V support).

After using printf function I have noticed that linker asks for some functions like _write which are used internally. I have added a file newlib_subs.c to implement these function.

You can download the example project from here.
The settings for usart communication are:
baudrate: 115200
parity: none
data bits: 8
stop bits: 1
flow control: none

If you want to use interrupts to read from USART you can use the code from here.

You can check my project via svn from my repository using:
svn checkout http://andrei-development.googlecode.com/svn/branches/dev/stm32f3-discovery


Saturday, November 17, 2012

STM32F3 Discovery + Eclipse + OpenOCD

Hi,

ST launched in September a very interesting development board(STM32F3-Discovery). It is a very cheap one(I have got myself one for ~10$). It has a debugger integrated(STLINK) and also some great sensors:
- ST MEMS LSM303DLHC, which contains 3 axis accelerometer(to measure acceleration intensity on each axis) and 3 axis magnetometer(to measure angles to a fixed point - the Earth's magnetic North)
- ST MEMS L3GD20, which has 3 axis gyrometer(to measure rotation speed)
This board is very good for automated pilot controller projects.

After unpacking the board I have found that it was supported just by commercial software and tools. As I am an opensource kind of guy I have struggled myself some time to get this working with Eclipse, OpenOCD and a free toolchain, on Linux.
I have used Ubuntu, but I think the process is the same on every distribution. Also, with little adjustments it can work on Windows.

Here are some steps, that you have to follow to get the led blinking example to work:
1. Install Java Runtime Environment. Here are some steps for Ubuntu:
sudo add-apt-repository ppa:webupd8team/java
sudo apt-get update
sudo apt-get install oracle-java8-installer

2. Install Eclipse. Get Eclipse IDE from here (grab the Eclipse IDE for C/C++ Developers) and unpack it somewhere

3. Install GDB Hardware Debugging. Open Eclipse go to Help->Install New Software and then search for GDB Hardware Debugging and install it.

   Install GNU ARM Eclipse plugin. Get it from here and install it from Help->Install New Software->Add->Archive and select the downloaded .zip file.

4. Install some dependencies. Paste following text in Terminal:
sudo apt-get install git zlib1g-dev libtool flex bison libgmp3-dev libmpfr-dev libncurses5-dev libmpc-dev autoconf texinfo build-essential libftdi-dev libusb-1.0.0-dev

5. Install OpenOCD(version>0.6.1). Get it from here and unpack it. Then, navigate to the extracted folder and type in Terminal:
./configure --enable-maintainer-mode --enable-stlink 
make 
sudo make install

6. Add rule for Stlink to be accessed without sudo. Type in Terminal:
sudo gedit /etc/udev/rules.d/99-stlink.rules
Paste the following text:
ATTRS{idVendor}=="0483", ATTRS{idProduct}=="3748", MODE="0666"
Type in Terminal:
sudo udevadm control --reload-rules
Now, with the board connected to PC you can enter  in Terminal:
openocd -f /usr/local/share/openocd/scripts/board/stm32f3discovery.cfg
The following text should appear:
Open On-Chip Debugger 0.6.1 (2012-10-29-22:02)
Licensed under GNU GPL v2
For bug reports, read
    http://openocd.sourceforge.net/doc/doxygen/bugs.html
adapter speed: 1000 kHz
srst_only separate srst_nogate srst_open_drain
Info : clock speed 1000 kHz
Info : stm32f3x.cpu: hardware has 6 breakpoints, 4 watchpoints
You can close the Terminal now.

7. Install toolchain.  You can use Codesourcery toolchain, like described in this step, or you can use GCC Arm Embedded toolchain(from here) which has support for hardware floating point unit and which is free and it is easier to install.
For Codesourcery follow the next steps:
Go to Codesourcery and download IA32 GNU/Linux Installer. To install it open the Terminal and navigate to the folder where it is downloaded. Type:
chmod a+x arm-2012.03-56-arm-none-eabi.bin 
./arm-2012.03-56-arm-none-eabi.bin
Then select next at every step. 

8. Download sample project stm32f3.tar.gz file and unpack it.

9. Open the project in Eclipse. Open Eclipse and go to Workbench. Click File->Import and then select General->Existing Projects into Workspace. Select the downloaded project and click finish.
This is a makefile project, so you have to edit makefile if you want to change some project settings. The frst thing you should do is open the Makefile file and check at the very beginning if the toolchain path is correct. This should be like:
TC = <path_to_toolchain>/arm-none-eabi
Now you can build. right click on project name and select build. The correct output is in the Console tab from Eclipse(ignore the warnings and errors from Problems tab).

10. Debugging. After the project was builded correctly select Run->External Tools->External Tools Configuration. Select OpenOCD(restart) in the left tab and click run.
Now, right click project and select Debug as->Debug Configuration and then
select in the left stm32f3-debug and then click Debug.
Note: If you want to add more source files you can add them in the src folder. New headers should be added into hdr folder. If you want to add another folder you have to specify it like the LIB_SRCS in the Makefile and also create LIB_OBJS like variable in the Makefile.

Happy free coding and debugging! :-)