Smart Laziness

How well do you sleep?

2010-06-11T15:52:00.000+02:00

Over the last few weeks, my friend Michał gave a little shape to our project VideoSleep. He wrote GUI in QT4 and enhanced reporting functionality I've rewritten motion detection using OpenCv and added some simple noise reduction. Before you will get hands on experience and whole code goes open source you can watch short screen-cast:

If you have some ideas of how you can contribute to our project, or want to try it out during your own sleep just leave a comment.

Under the hood

2010-05-31T23:43:00.000+02:00

Logger
Obvious part, every error and finished stage of migration should be clearly logged. Then it will be used when displaying information and progress of import in scilab batch. Scilab for logging uses apache commons logging, so no choice left here.
Parser
There is already very convenient way of parsing Simulink model files. Conqat is an open source project (licensed under Apache License 2.0) that provides Java's Simulink parser library. It's well documented and I think it should be the best existing solution. Sample usage of this parser can look like this:

SimulinkModelBuilder builder 
 = new SimulinkModelBuilder(new"), new SimpleLogger());
SimulinkModel model = builder.buildModel();

then every block, parameter can be accessed by simply:

public static void readSimulinkBlock(SimulinkBlock block) {
  UnmodifiableIterator blockIter =       block.getSubBlocks().iterator();
  while(blockIter.hasNext()) {
       readSimulinkBlock(blockIter.next());
       }
    }

Below, the class diagram (of Conqat Simulink Model) that clearly shows how Simulink models should be processed:

Mapper
This step is the most demanding in whole idea of creating 'Simulink Import'. It requires implementing compatibility patterns for evaluation functions and block representations (and there are a lot functions and block in simulink). In this part I'll create some clean XML migration schema and process it using JAXB (Java Architecture for XML Binding).
All mapping will be conducted directly. Recursively accessed blocks (every block can be subsystem) will be decoded straight (just like in io.scicos) to XcosDiagram via binded XML migration schema. Every element of Simulink diagram will be checked during decoding (in case of troubles proper logs should be made).

We can make possible to manually correct the diagram after unsuccessful importation, that is why giving good feedback from mapper is even more important. After this stage of import we should have existing XcosDiagram structure.

Simulink import

2010-05-25T19:19:00.000+02:00

Why?
At this time migrating Xcos scheme to Simulink is almost impossible. It requires the user to create entire diagram from scratch. It is unacceptable for many reasons, such as : great amount of time needed, required knowledge of both Simulink and Xcos, required license for Matlab/Simulink during whole (sometimes very long) migration process.

Look and feel
Simulink import will be available to users via batch functions and menu entries.

sim2xcos('filename.mdl');
Translates and loads diagram from mdl file to new Xcos window.

sim2xcos('simulink file path', 'result path');
Translates diagram from .mdl file to .xcos file.

sim2xcos();
Opens window with file selection, after file selection model is imported and opened in new Xcos window.

scs_m = sim2xcos('filename.mdl');
Creates xcos_diagram from .mdl file.

It can be also good idea to add 'display information mode' choice with verbose parameter.

Simulink import will be also available as 'File menu' entry in Xcos:

Selecting 'Import...' from menu will open import dialog:

After selecting file with model, imported diagram will be opened in current Xcos window.

It would be great to add 'Simulink import' to 'Matlab to Scilab translator'. 'Simulink import' will have to import every model called by matlab 'sim' function, and then run it with scicos_sim() and proper parameters.

Next part with 'Under the hood' coming soon...

I've been accepted to GSoC 2010

2010-04-30T14:41:00.000+02:00

So the fun begins, over a few next weeks I'm going to work on my university projects, learn for exams, bound with Scilab community etc. But when 24th may comes I've got to be ready!

Yesterday morning I had first surprise concerning GSoC. The gsoc-students-mailing-list. It's hard to believe that all of these people got accepted. How will they get on with their mentors knowing nothing about netiquette. Over a 130 e-mail messages, and 16 new threads filled with pure spam. I can't resist to quote my favorite message here:

Hi,
As We are student developers selected in a Google Event.Shouldn't program administrators ensure that Google search of our names give a Page with High Rank.
For Example by putting a list of selected students on a www.google.com domain page?
This can highly motivate students and particiaption in future events.

Fortunately we have Michał (one of accepted students from Poland). Who cast a spell of spam transition:

If you enjoy using IRC (which by the way is very suitable to make real time introductions ;) ), you can try joining #gsoc-XX at irc.freenode.net, where XX is a two letter country code of the country you reside in (or the country you are from). You can check your country code here http://en.wikipedia.org/wiki/Country_code .

It made mailing list usable again, what's more you can now join #gsoc-pl to discuss with accepted students from poland.

Soon I'll try to write something about submitting a bug patch for Scilab, update my progress on all projects, maybe even show you my GSoC proposal. Stay tuned.

Simple motion detection in Simulink.

2010-04-16T19:52:00.000+02:00

Motion detection will be an important part of our video sleep analysis package. As the studies (1) show pure amount of movement (without recognition of individual parts of body) can be sufficient to recognize legs or arms motion. What's more, with some further analysis I hope I'll be able to distinguish respiratory waveform in the quantity of movement over time graph.
So let's have a look at the simulink diagram I've created:

I don't really have much video footage of sleeping people for now (but I'm going to record some when I get appropriate camera with night vision support). So to show you how my motion detection works I asked my friend and co-author of the Video Sleep package Michał to lie on the table for a while. Here are some results of our experiment:

And below, plot of movement over time:

As you can see on the plot, every Michał move is registered, during whole recording he was moving and breathing as normally as he is able to :) but between 20 and 33 second he held his breath unnaturally (during sleep it may be a sympthom of sleep apnea).

If you are interested in the topic of video sleep analysis it's worth to look at:
(1) A Monitor for Posture Changes and Respiration in Bed Using Real Time Image Sequence Analysis
Kazuki Nakajima, Yoshiaki Matsumoto, Toshiyo Tamura

Scilab compilation under Archlinux

2010-04-06T15:09:00.000+02:00

Last weekend, after a few hours I've finally managed to compile Scilab in my Arch. After the break I described all the problems I've encountered.

I started with installing dependencies. All of them but PVM can be installed via pacman. To install PVM and to be able to make scilab without any -fPIC error you have to download PVM sources from http://www.csm.ornl.gov/pvm/, add following lines to your .bashrc:

export PVM_ROOT=/home/xendo/GSoC/pvm3 # location of pvm3 sources

export PVM_ARCH=`$PVM_ROOT/lib/pvmgetarch`

export PATH=$PVM_ROOT/lib:${PATH}

export PATH=$PVM_ROOT/lib/$PVM_ARCH:${PATH}

export PVM_RSH=/usr/bin/ssh

then simply make pvm with 'shared' parameter.
Next I downloaded scilab sources, and prerequirements:

git clone git://git.scilab.org/scilab scilab.sources

svn export --force --username anonymous --password Scilab svn://svn.scilab.org/scilab/trunk/Dev-Tools/SE/Prerequirements/Linux/bin scilab.sources/scilab/bin

svn export --force --username anonymous --password Scilab svn://svn.scilab.org/scilab/trunk/Dev-Tools/SE/Prerequirements/Linux/thirdparty scilab.sources/scilab/thirdparty

When I tried to ./configure I got following error:

checking for h5JNIFatalError in -ljhdf5... no

configure: error: libjhdf5: Library missing (Cannot find symbol h5JNIFatalError). Check if libjhdf5 - C/Java (JNI) interface for HDF5 - is installed and if the version is correct. Note that you might have to update etc/librarypath.xml to provide the actual path the the JNI libraries.

To avoid this download binaries from scilab download page and then copy *hdf5* files from scilab_bin/lib/thirdparty to scilab_source/bin. After this you will be able to ./configure and make all.

Next error showed up during make install, and it looked like this:

/bin/install: cannot stat `./examples/Project/Settings.Designer.vb': No such file or directory

/bin/install -c -m 644 ./examples//call_scilab/NET/VB.NET/My xsci/modules/call_scilab/examples//call_scilab/NET/VB.NET

/bin/install: cannot stat `./examples//call_scilab/NET/VB.NET/My': No such file or directory

/bin/install -c -m 644 ./examples/Project/Settings.settings xsci/modules/call_scilab/examples/Project

Solution to this one is kind of weird, you have to copy all files from:

scilab.sources/scilab/modules/call_scilab/examples/call_scilab/NET/VB.NET/My Project

to two other locations:

scilab.sources/scilab/modules/call_scilab/examples/call_scilab/NET/VB.NET/My

scilab.sources/scilab/modules/call_scilab/examples/Project

This one took me a while to crack. And I wonder how it's possible that it hadn't showed up under ubuntu (if you know that, leave a comment). After this I was able to successfully finish Scilab compilation process.

Running pacman from LiveCD

2010-04-04T14:45:00.000+02:00

Not every Arch user knows that you can run pacman with an alternative installation root. It's very useful especially when you are using [testing] repo, while not reading carefully arch-dev-public mailing list. After a simple upgrade you can get something like this:

error while loading shared libraries: libssl.so.0.9.8: cannot open shared object file

(topic from bbs about this error) and find out that neither your X nor pacman is working.
But don't panic! Enter your Arch LiveCD, log in as root, mount your system partition, disable testing repo, and run pacman do downgrade nonworking packages:

fdisk -l
mount /dev/sda? /mnt/mountpoint

vi /mnt/mountpoint/etc/pacman.conf #comment testing repo for a while
pacman -S pacman -r /mnt/mountpoint

How easy is to translate open source project?

2010-04-02T00:45:00.000+02:00

Nie wiem ilu z was próbowało kiedyś tłumaczyć jakiś projekt open source. Wiem natomiast, że większość z tych, którzy nie próbowali byłoby zaskoczonym jak łatwo można to robić. Wszystko dzięki launchpadowi - aplikacji sieciowej stworzonej przez Canonical w celu ułatwienia utrzymywania projektów z otwartym kodem źródłowym. Przejdźmy może do rzeczy. Za przykład wezmę Scilaba jako, że mam nadzieje, że któryś z mentorów tegorocznego GSoC zajrzy na tego bloga i uzna, że to miło z mojej strony, że to akurat Scilaba promuje ;)

Na chwilę obecną w samym Scilabie jest do przetłumaczenia około 3200 takich zwrotów, więc do dzieła wszyscy amatorzy dobrego tłumaczenia. A tak btw to słyszałem, że tłumaczenie projektów open source dobrze wygląda w CV!

Ideas of development for Scilab

2010-03-28T14:15:00.000+02:00

Simulink import is one of the most interesting projects from GSoC list this year. It's challenging development problem:

- simulink file parser should be able to handle .mdl format changes,

- xcos and simulink function blocks, solvers and all other parameters differ in many ways and may differ even more in the future. It would be great if "simulink import" could adapt to possible changes,

- xcos lacks a lot of simulink functionality, this project can give scilab community feedback which simulink functionality users miss the most.

From the other side, with improved Matlab to Scilab conversion tools and ability to import simulink files it will be possible to easily migrate Matlab projects with Simulink simulations to Scilab/Xcos. This cannot be overestimated!

Computer Vision toolbox for Xcos is my own idea which I'll try to develop after GSoC. Visual programming is awesome, and 'Computer Vision' was one of my favorite courses, so bringing this two together sounds like a great idea. Implementing SIVP functionality in Xcos will be good thing to start with, adding GpuCV abilities can be next step, and finally some FPGA support (like System Generator) may be a great idea for my Master of Science thesis.

GSoC

2010-03-19T18:43:00.000+01:00

I've decided to describe my way of becoming Google Summer of Code participant in 2010. I'm a student of Automatics and Robotics which is probably why I decided to apply, to one of the most engineer 'friendly' open source projects: www.scilab.org. About a month ago I subscribed to Scilab mailing lists and I started visiting their irc channel, I've also carefully studied list of scilab development ideas.

P.S.

Yesterday the list of accepted mentoring organizations showed up on GSoC page, congratulations Scilab!