scanning humans

[cees10]

Well called sick today we had a long talk  the could not handle the speed and resolution

I want 1 mm at all planes  about 1,5 meter long  is the scaned body  athelte in his postion

and what i want to have it scanned in about 2   to 3 min    becuase a athlete cannot sit so long absolute still.

[Hummer1]

I still think that you could do a really good job using a stationary trainer and video from a couple of directions.

Hummer1

(Guilty of staying in the first stage of the 5 stages of a project...)

[cees10]

Again i can video it's  no problem, but with a video i do not get data points  to create a solid ,

[wd8ivl]

There was a video on TV a couple of years back that showed how they did it with Lance Armstrong and Trek for the TDF.  I think it was on the old OLN (now Verses) sports TV network.

If I remember, they attached some florecent dots at his joints and used a camera and filtered to just get the color of the dots - then video captured the moving dots with a PC.

Maybe Google UTUBE for the video?

Sounds like a fun project.

Good luck,

Mike   

[Hummer1]

The vision tools can find the outline of the rider and bike...probably seperating the bike and rider if you have them different colors.  Background should be a distinct color from either.  Use the vision tools to find the object border of the rider in each frame and then fill in the form...by doing it from at least two directions, (perhaps adding the dots described above as location reference points), I think that the outline in one dimention, could be mapped to the outline in the other direction and combined to give a volume.  Ues two cameras close together at each location and add the "stereo" distance measuring capability...again, might be nice to have dots or a grid on the suit to find corresponding points in the images from the different perspectives. 

Nice thing about video, once you get the subject "strapped up" there is no need to process the shapes very fast, just take video and do some simple calculations to make sure you are able to "see" properly.

I recall seeing some functions that are matlab based for processing images that will pick "important" points for multiple image registration. 

here is a web discussion from Wright State that I found on the web...but the general topic of point registration is what you want to persue.

http://www.cs.wright.edu/~agoshtas/CVPR04_Registration_Tutorial.html

Hummer1

Good Luck

[rex1030]

Can you tell us more about your application?

What you are asking for (fast, accurate LADAR) is not cheap. Rather than LADAR producing a massive point cloud that needs to be handled and analyzed, I would look into the use of multiple cameras that produce images that are analyzed to produce a 3-d model. This is usually called "stereo image analysis". There are a lot of thesis papers written on it you may be able to get your hands on. Also there is software out there to do it as well. Try to look for analysis software that uses neural network algorithms because they perform better and faster than traditional analysis methods.

I did a project on terrain analysis in my Advanced Neural Networks course last year and did a lot of research on it. I would try to find some out-of-the-box software rather than develop your own method because it can be pretty complicated. However, it can be very effective. As you will see, people do graduate projects on it.

[cees10]

wd8ivi  

Its about the same as lance

As we went it to windtunnel we also did before  scan with FARO scanner   well really lot of problems to make a solid and time to scan was too long.

(and yes we created a cloud but this is not a way to go in real life) 

With a camera is no solution it hard to capture through vision human body tried  it ,  i calc live frontal aera trough vision  in windtunnel.

The dots must be so small!  to create a good solid!  

so it really must be good and fast 

rex1030  

application  calc drag of bike racer (so we need the real solid)