Julien's fuzzy blog

Today was my first UbiSense meeting. I had to present a couple of slides about the simulator I've started to develop. It went ok.

Most interesting was to see the relationship between the companies and the universities involved in the project. Not exactly the same approach and aims ...

More than 700 photos have been taken during these 3 days by Surapa (on the third picture). Pictures are now available on the UbiMon website. Many thanks to her !

   

Today was cool. A proper conference day, with coffees every 2 hours . Not much to worry about. Only informal demos of Rachel's application (to look clever) and mine at every tea break. I could attend some talks, that was quite interesting. I particularly liked a speaker that was ironically expecting a "smart condom that would remember you how to use it"

Hey ! The demo went eventually ok ! Actually somebody that I don't have the right to name here just found out that there were some network sockets available in the demonstration/poster room !
So I could run my demo without any modifications. Was not really impressive, but people were quite impressed it was done in only 1.5 months. Also lots of them laught cause of the cow...

Hope that tomorrow, I will be able to follow some talks instead of ./configure ; gmake ; gmake install ; urpmi.

Yesterday I learnt that it was impossible to create a local account on a linux machine here, because NFS is stronlgy integrated to the configuration. And this afternoon, I have to give a demo of my application in a room without network. The only solution was to configure a new machine by myself, from scratch, without network, keyboard, mouse or screen. I spent yesterday afternoon and this morning switching between these two computers, plugging in and out cables, rebooting every five minutes to install the required packages. And I finally surrendered. I will present a simple slide show I think .

In the mean time, we have to welcome and help people arriving to the conference. This is quite nice to be on the other side...

As you've read my blog a couple of weeks ago, you already know that the group is organising the International Workshop on Wearable & Implantable Body Sensor Networks next week. Because of that, the lab gets quite busy at the moment. We have a couple of small jobs to do. A couple of days ago, we filled in 150 briefcases with the proceedings, tube and college maps, BSN pens, etc. Today was a bit more technical, as we prepared the "development kits", small boxes with BSN components. Cutting and pasting foam, soldering, contacts checking, label pasting...

Hey ! I'm a proprer registered student (well I hope so, I've not received the firm offer letter yet ) ! I've got a swipe card and new student pack.

void* Julien1 ;
sleep(1000*3600*24*60) ;
PhDStudent* Julien2 = dynamic_cast<PhDStudent*>(Julien1) ;

I'm about to finish my pre-project... and therefore to start my PhD ! I got exactly the subject I wanted: IBMR for surgery.

"Image-Based Modeling and Rendering" is the field of research that uses images from a set of cameras to build a 3D model of an object and even a 3D rendering. This is a really powerful technique to generate photo-realistic 3D scenes. A couple of links about that:
http://www-2.cs.cmu.edu/~ph/869/www/misc.html
http://peter-oel.de/ibmr-focus/
http://www.ibmr.net/

By the way, the weather is nice again today !

I've started to develop a filter plugin for the simulator that provides a kind of Night Vision Device (NVD) view. These devices are based on electron acceleration/multiplication, and are frequently shown in movies such as "The Silence of the Lambs".

 

It is simply based on a blurring effect, a shift to the green and a kind of saturation filter.

Image not clickable

Last week I designed a poster on what I've done so far (unfortunately the filter plugin was not available yet).Today it has been printed out on a high quality A0 sheet of "Kodak Royal Paper", which looks more like PVC than paper. A royal paper is the minimum required for the Imperial College . I will present it at the International Workshop on Wearable & Implantable Body Sensor Networks in mid-April.

As you know, I work on a camera network simulator. The software I'm building is now able to load through XML files:

• the 3D model of a building
• a camera network setup (position, angle, etc)
• a programmed path for a person (here a cow )

Moreover, I've developed a system of plugins. The developers can compile their plugins separately as shared objects (*.so/*.dll). Such plugins can then be dynamically loaded inline.
An interesting type of plugin is the filter, allowing to process operations on the output of a virtual camera. They'll be used to test gait detection algorithms for example. You can plug in as many filters as you need, building a tree of filters.

Screen shots are more talkative . First one is an overview of the system:



The 4 following detailed views represent:

• the original view
• a blur filter applied to this view
• a red filter on the output of the blur filter (to test filter chain system)
• a movement detection filter (to test temporal aspect)

Of course, these filters are not useful at all, I've done them only for test purpose.

       

The application relies on:

• OpenSceneGraph
• Open Inventor (for osgPlugin - VRML)
• libXML
• CImg for image display (it's a shame, this French lib can do far more !)

What's next ?

• Debbugging .
• All the objects (building, cow, camera, filter) can be loaded from a console, which is not convenient. I will ameliorate it.
• A WxWidget-based GUI to setup the tree of filters.
• SQL data loader
• Python plugin system for fast prototyping
• More interaction in the OpenGL viewer (hide objects, etc)

The previous posts about my work here were quite narrow-minded, focussing on particular aspects. Hereafter a better overview.

This project is actually part of a larger scale project whose aims are numerous. One of those would be to profile the behaviour of people at home. This relies on a camera network which processes the images inline.

The part I am devoted to is a house simulator. This simulator should embed:

• a 3D model of the building to test a project before being build
• a camera network model to maximize the visibility field, evaluate the image processing time and simulate the computing power transfer between these devices
• a human model to test some behaviour recognition algorithms

I've just achieved the first goal of this pre-project: I've implemented a "visibility pyramid" method in order to detect whether or not an object is visible to the camera (based on the outline of previous coders).

The already implemented algorithm was based on classic ray-casting, that consists of sending rays from the camera to the object and detect a potential collision. The number of rays is obviously finite (this represents the resolution of the camera).

The algorithm I've implemented is completely different. It consists of creating pyramids which apex is the center of the camera and which base is a face of the model to track. If the face is partially obstructed by something, then the pyramid is cut to keep only the visible part of the object face. A recursive cutting of all the obstructed parts of the face is proceeded. Eventually the pyramid is either null and therefore the object is not visible or not null in which case the object is at least partially visible.

Is that clear ???

Today I started to code seriously. I base my code upon a Java project. The developer is excellent, but put a very few comments. 6 lines of comments at the beginning of a 200-line method... then nothing ! Moreover, the average variable name contains 4 characters, and if longer look anyway like "vdPthyInc" (real example). Wow !
Hopefully, this guy is just really kind and helpful, and answer to every question.
The other point is that I can't use libraries such as Java3D or OpenSceneGraph. So Im writing for the tenth time of my short programmer's life the vector cross product function, etc...

I've arrived in London yesterday and I'm now living in a hotel near Victoria for 3 days. Looks cosy and weird at the same time to my French eyes, like most of that kind of 50-year-old British houses. Bible and kettle in every room. Then I should stay one week (maybe more) in a guest house near Hyde Park. This gives me at least 10 days to find a proper accomodation.
I still don't know exactly where to search for, as each person gave me different advice. Near the campus or not ? In an university hall or in the private sector ? Some told me it's easy to find, others not at all...


I've started to work on a mini-project that should last 2 months till the official beginning of the PhD. It's not directly related to medical imaging: it's about visibility detection. Given a 3D environment (eg. a house), a 3D model (eg. a human being) and a camera, the application should evaluate the visibility of the model in the environment by the camera (ie. basically if the model is in the camera field of view and not completely hidden by other objects). This could lead to an automated power on/off system for a set of cameras, in order to record video only when the considered model is visible.
I should base my application upon an undergraduate group's project coded in Java, but I'm already hesitating about starting back from scratch.

Hope I will meet my 'Londoner' friends soon and Hull's ones in the coming months.