No announcement yet.

State-of-the-Art in Markerless Motion Tracking

This topic is closed.
  • Filter
  • Time
  • Show
Clear All
new posts

  • State-of-the-Art in Markerless Motion Tracking

    Hi everybody,

    I am by no means an expert on markerless motion capture, but like many
    of you I am intrigued by its potential. I am enough of an "old-timer"
    to have experienced several landmark steps in motion capture:

    1. Going from film to video

    2. 2D to 3D

    3. Manual to Automatic digitizing

    4. Real-time display

    Of course, along the way, there have been huge improvements in
    resolution, frequency, animation, and computation.

    This is an exciting time in biomechanics, because we are on the verge of
    the next landmark step - automated, markerless motion capture. Tom
    Andriacchi, Ph.D., has been a leader in introducing this to most of us
    ( There are now
    several others working on this too. Two that I know about are mamoca
    ( and Organic Motion (

    A few weeks ago, there were a series of postings regarding automated
    markerless motion capture, particularly with technology from Organic
    Motion. Below is the final posting on that series - a response from
    Organic Motion. I had an opportunity to meet with Organic Motion last
    week, so I thought I would share some observations here.

    First, the executive team at Organic Motion seemed like good, solid
    people. They explained where their company is and where they are trying
    to get to. Then, we discussed briefly how their technology works.
    Finally, they gave me a demonstration. The demo was a calibrated space
    - perhaps 15ft x 15ft - with reflective white walls and floor. There
    were about 14 small cameras around the edge of the space, each with a
    ring of visible-light LEDs surrounding the lens. I walked into the
    space, and instantly I saw my real-time animation on a computer monitor.
    I performed some simple tasks - waving, jumping - and it tracked my
    motion. They collected a few trials, and we reviewed some simple
    kinematics (elbow flexion vs. time). While I could not assess the data
    accuracy in this brief visit, I was impressed with the real-time, 3D,
    markerless data capture. There seem to be some challenges ahead for
    Organic Motion - namely, validation of the accuracy and also making the
    data capture more robust for bigger, more complicated motions. However,
    Organic Motion showed me that they have solved the initial hurdle of
    markerless tracking, so they are clearly moving in the right direction.

    - Glenn S. Fleisig, Ph.D.

    Glenn S. Fleisig, Ph.D., Smith & Nephew Chair of Research
    American Sports Medicine Institute
    833 St. Vincent's Drive, Suite 100
    Birmingham, AL 35205
    (tel) 205-918-2139

    -----Original Message-----
    From: Glenn Fleisig
    Sent: Monday, April 21, 2008 4:14 PM
    To: Glenn Fleisig
    Subject: Organic Motion

    -----Original Message-----

    From: * Biomechanics and Movement Science listserver
    [mailto:BIOMCH-L@NIC.SURFNET.NL] On Behalf Of Alex Czarowicz

    Sent: Wednesday, March 19, 2008 6:38 PM


    Subject: Re: [BIOMCH-L] State of the Art in Markerless Motion Tracking

    This is Alex Czarowicz, VP at Organic Motion - posting the official

    Organic Motion response to questions and statements about our computer

    vision / markerless 3D motion tracking technology.

    First we want to thank all of you who have taken the time to post your

    comments. Some of these posts did include a few inaccuracies about our

    technology, and we wanted to join the discussion and 'set the record


    Below we outline how BioStageTM - Organic Motion's solution for the Life

    Sciences works from a technical perspective, and address the validation

    process and other ongoing developments.


    An Overview of BioStage - how it works:

    BioStage breaks away from the pattern comparison approach used by all

    optical marker-based systems. Instead BioStage actually 'sees' the

    entire person and tracks thousands of points on each subject. BioStage

    recognizes human shapes and can analyze precise activity without any

    need for attached 'cheating' devices (markers). We do not have a

    library of movements that helps build our tracking model; instead

    BioStage tracks a person's actual full body motion in real-time.

    BioStage uses 14 monochrome 2D cameras to view the subject and

    incorporates three general processes to create an accurate 3D model of

    the subject in real-time:

    Video Sub-System

    The Video Sub-System acquires lens and space calibrated video from the

    cameras and delivers these synchronized streams to the reconstruction


    3D Reconstruction System

    The 3D Reconstruction System turns the 2D video streams into 3D point

    and surface clouds by triangulating the various 2D viewpoints. In this

    way the 3D Reconstruction System acts much like a 3D scanner. This

    system generates massive amounts of data that resembles the human

    subjects as they appear inside the scan space.

    Character Fitting System

    The last step involves "recognizing" the human figure in this 3D data

    cloud. Here BioStage uses a complex rules based approach which maps a 3D

    humanoid skeleton into the data. The output data BioStage delivers is

    the X, Y and Z positions and orientation (6 DOF) of 21 segments of this

    skeleton and is then loaded in real-time through a plug-in directly into

    biomechanical processing software (The MotionMonitor from Innovative

    Sports Training or Visual3D from C-Motion). In addition a full 3D mesh

    image, complete with surface textures is provided.

    Currently the Character Fitting System is specialized to recognize and

    track a typical human biped.



    The capture volume can range from about 4 by 4 feet, up to approx 16 by

    16 feet, and is flexible in its way to be set up for different analysis

    needs/tasks in a lab. The current version of BioStage includes a

    reflective white cloth backdrop which makes it easier for BioStage to

    process the video data. A color camera system will make BioStage more

    flexible to scan subjects without the need for the reflective backdrop.

    Organic Motion will implement the color camera system within the year.

    Data formats:

    BioStage generates data that flows real-time into both The MotionMonitor

    from Innovative Sports Technology and Visual 3D from C-Motion. Organic

    Motion also provides an SDK to access the motion data directly.


    Ongoing Development & Validation:

    Organic Motion is in the process of performing validations of BioStage

    for the Life Science community, with the expectation to demonstrate its

    tracking accuracy. We are currently involved in several validation

    projects, and are seeking one or two more validation studies at this


    The requirements/goals of these studies are to:

    a. provide a fitting system that ensures precise, repeatable, consistent

    length measurements of human bones/segments.

    b. deliver anthropometric measurements through a defined protocol in a

    totally "non-invasive" method in real-time.

    c. evaluate BioStage, its performance, and to assess its accuracy,

    reliability and suitability for use in research, sports and clinical


    d. perform an assessment of errors that affect temporal, kinematic, and

    kinetic variables when estimated by means of the proposed markerless

    system compared to the existing markerbased motion capture systems using

    reflective markers currently used in motion analysis.

    e. validate BioStage for biomechanical research, sports and clinical



    I will be glad to show some of the research done so far on an individual

    basis with typical confidentiality rights in place.

    For further questions/comments please contact me at

    Thank you.

    Alex Czarowicz


    Information about BIOMCH-L: