3-Man Basketball

Inputs (two separate events)
Movie
First event snapshot Second event snapshot
Reconstructed 3D shape
Movie Timelapse image
Start time Intermediate time End Time

Virtual View of combined event
Movie
Rendered Virtual view
 
What are we doing?
This example demonstrates many of the capabilities of the Virtualized RealityTM approach to dynamic event modeling. Two dynamic events are captured separately, and digitized as a sequence of 3-D models. Both these events are combined with each other by applying suitable geometric transformations to each of the sets of 3-D models. The 3-D models are then enclosed within a CAD model of a basketball court, and are used for synthesizing new views.  Note that this is completely different from bluescreening.  

More Flythrough results
 

  • circle around as the basketball moves... (363 K MPEG)
  • Linear moving view point (395K MPEG)
  • Fly into the gym (1,063K MPEG)

    • How two events are combined:

    2-Man Basketball event

    The first event contains two people playing basketball. One player has the ball, dribbles, and makes a finger-tip pass. The other player is trying to block the pass and to steal the ball. The dynamic virtual modeling process generates a sequence of 3D models (1.5 MBytes, gzip'ed Open Inventor). An MPEG movie (288K) of the models is available if you can not view the models themselves.) A static 3-D model (101K, gzip'ed VRML) of one frame in the sequence is also available.

    1-Man Basketball event

    The second event consists of one person catching a basketball and dribbling. The modeling process generated a sequence of 3D models (1.33 MBytes, gzip'ed Open Inventor). An MPEG movie (399K) of the models is available if you can not view the models themselves.) A static 3-D model (55K, gzip'ed VRML) of one frame in the sequence is also available.

    Mixing the two in 3-D

    Once we have recovered the dynamic models, the two events can be combined. In this case, we want the pass in the 2-man sequence to connect with the pass in the 1-man sequence. We move the model of the 1-man player to align with the flight of the ball from the 2-man sequence. We then concatenate the two events in time, so that first the 2-man sequence plays, and then the 1-man sequence plays. In 3D, though, we want the players to co-exist, so we can't allow the models to just appear and disappear. Instead, we "freeze" the ends of the two sequences so that they players appear throughout the event. In these freeze frames, we also eliminated the ball so that it would not appear twice. The resulting model sequence can be seen as an MPEG movie (1.42 MBytes). A still frame from this sequence is available as here (23K JPEG ), while the entire  and as a 3D VRML file (154K, gzip'ed VRML).

    By super-imposing the combined event over time, we get an image similar to a time-lapse photograph (54K JPEG ) of the combined event, which shows the amount of scene motion over time. We can also show the motion as a movie (385K MPEG) of progressively longer time-lapse photographs.

    Return to the Virtualized RealityTM Home Page 
    Peter Rander, Sundar Vedula