Our system can dynamically
recalibrate the cameras
in the view cube.
Our approach to calibration is to maximize the consistency between image or alternatively the reconstructed volume. An outline of the algorithm:
Our system can uses a graph cut
formalism to find the optimal
segmentation very rapidly.
Our system operates in real-time. Each camera computes a segmentation and then compresses the silhouette by run-length encoding.
The compression depends on the orientation of the camera in the world. As a result 3D reconstruction of the volume can take place without decompressing the silhouettes.
with graph cuts
Our system operates uses the graph-cut formalism to find the optimal 3D volume which explains the resulting silhouettes.
The graph-cut system incorporates both a spatial prior (on shapes) and a model of noise in the images.
|Back to first slide||View graphic version|