PDA

View Full Version : Re: State of the Art in Markerless Motion Tracking



Unzueta, Luis
03-16-2008, 07:20 PM
Hello everybody. I neither had direct experience with the Organic Motion system, but looking at the videos of their webpage it can be seen that they don't just get the surface of the user and render it along with its textures in 3D. They are also able to reconstruct the 3D skeleton of the user in real time. The movements of the 3D skeletons they show running in MotionBuilder are very smooth and the system seems to capture also movements of the pelvis and shoulders (bearing in mind the simplification of the shoulder complex they do) very realistically. Maybe these videos have been postprocessed, but it's supposed they are not. The question to know for its applicability in Biomechanics is the precision in the subject calibration. They just say that:

* XYZ Positional Accuracy (arms, legs) up to 1.0mm
* Rotational Accuracy (arms, legs other than along
axis) up to 1°
* Rotational Accuracy (along axis) up to 2°
(along axis)
* Actor Calibration: On site in under 1
minute - Stage auto-learns after first
'acquisition' of Actor

Which is not a very detailed description of the reconstruction's accuracy.

Apart from this system I would like to point out these two interesting markerless approaches:

http://markerless.com/company.html
http://www.softkinetic.net/Public/eng/

The former seems to be very similar to the Organic Motion's approach. The later is more oriented to Human-Machine Interaction than for Biomechanics but it also allows the reconstruction of the user's poses and is a good example of the use of 3D cameras.

Finally I would also like to remind that the International Journal of Computer Vision (IJCV) will publish (deadline for submission just passed on January 11) a special issue on Evaluation of Articulated Human Motion and Pose Estimation (http://www.editorialmanager.com/visi/). The accuracy of the presented methods will be measured with the HumanEva database (http://vision.cs.brown.edu/humaneva/index.html).

Best regards

________________________________________
Luis Unzueta Irurtia
Centro de Estudios e Investigaciones
Técnicas de Gipuzkoa, CEIT
www.ceit.es
Tfno: (+34) 943 212 800 Ext. : 2246

-----Original Message-----
From: * Biomechanics and Movement Science listserver [mailto:BIOMCH-L@NIC.SURFNET.NL] On Behalf Of Stefano Corazza
Sent: sábado, 15 de marzo de 2008 18:41
To: BIOMCH-L@NIC.SURFNET.NL
Subject: [BIOMCH-L] State of the Art in Markerless Motion Tracking

While I don't have direct experience with the Organic motion system,
from various articles and the material on the website it seems that it
still has some practical limitation. Currently the system uses black
and white cameras and requires a white screen to perform background
subtraction (
http://www.wired.com/gaming/gamingreviews/multimedia/2008/02/gdc_gallery_2?slide=2&slideView=2 ). There doesn?t appear to be any quantitative validation data currently available. In addition, the system operates as a black box and it appears it would be difficult for biomechanical applications to access variables like the definition of the anatomical coordinate frames and joint center positions. Given these limitation its capacity to reconstruction kinematics in real time is
impressive.

In a biomechanical analysis context, it is important to notice that
the two fundamental approaches in markerless motion capture have
different sensitivity to the skin artifact phenomenon. Simplifying a
bit they can be defined as:

1. Surface Feature Tracking: This approach is based on tracking local
features on the subject's surface,
2. Volumetric Reconstruction: This approach is based on reconstructing
the subject's volume from silhouettes and tracks the reconstruction
with an a priori defined model.

The volumetric approach, due to its different formulation, is not as
sensitive to local skin artifact as marker based systems. However, the
approach is quite sensitive to camera resolution and the number of
cameras.

At the Stanford Biomotion Lab we have developed a markerless motion
capture system that uses the volumetric approach and have found good
agreement with marker based kinematic measurements. The current system
consisting of 8 VGA cameras has demonstrated the capacity to be used
effectively for biomechanical applications, not requiring an ad hoc
background and suitable for outdoor applications too.

For additional information see publications at
www.stanford.edu/~stefanoc/Markerless/Markerless.html

Best regards,

Stefano Corazza


Quoting Matt O'Brien :

> Along with 5 or so other manufactures of motion capture equipment, I saw the
> Organic Motion system at the 2008 Game Developer's Conference in San
> Francisco. Other companies such as Moven (a suit based system) and
> Optitrack (low end, $5,000 system) were there, along with the old stalwarts,
> Motion Analysis Inc.
>
> With the Organic Motion system, the capture volume is small, and permanently
> fixed. Apparently the company needs to come out and do the installation for
> you. As far as using it, it was pretty neat. After standing in a T-pose
> for one or two seconds, the sensors look for a difference between you and
> the white backdrop. Once the software gets a feel for you, you are free to
> move, without any sensors attached.
>
> According to the sales person at the booth, the software has a library of
> human movements with which it will try to validate to what it tracks from
> the human mover. Possibly a big red flag for research oriented users...?
>
> Overall, I didn't get the feeling like the system was ready for use in human
> movement or biomechanics right out of the box. I don't know how much
> tooling and customization it would take, and of course that really depends
> on what type of research the user is doing. It does, however, seem like a
> great solution for animation (at least for a small volume. Not sure about
> facial captures though). The company is only 2 years old, and from what I
> understand, they have sold their first few units this year to some animation
> studios in the UK.
>
> At the very least, it is great to see novel motion capture equipment come to
> market. The biomechanics applications still appear to be lagging, however,
> in most of the solutions. I assume most growth in this industry will
> continue to become more and more driven by applications such as animation,
> video games, and commercials.
>
> Matt O'Brien, B.S. Kinesiology
> San Francisco CA
>
> ---------------------------------------------------------------
> Information about BIOMCH-L: http://www.Biomch-L.org
> Archives: http://listserv.surfnet.nl/archives/Biomch-L.html
> ---------------------------------------------------------------
>