The Goal: Automatically
construct spline surface models from range data or photos.
We want to capture complex shapes
for vision
and graphics applications.
Representation:
quartic triangular Bezier patches
G1 continuous surfaces
arbitrary topology models
few parameters
Modeling from Range Data:
register range views
sparse data, unknown topology
approximate with canonical surface
iteratively fit to data
subdivide near high error
refit until tolerance reached
Range Fitting Results
Modeling from Silhouettes:
capture images with calibration setup
extract object silhouettes
build an approximating polyhedron
triangulate/simplify to form a control mesh
construct an interpolating spline surface
fit the surface to the silhouettes
texture map from calibrated photos
More Information
Spline Fitting Results
Pose Estimation from Silhouettes:
take image from arbitrary view
extract the silhouette
estimate initial pose
move camera to fit model
Pose Results
|
|
|
