Structure and Motion Estimation from Dynamic Silhouettes
under Perspective Projection

Estimation of the structure and motion of a smooth curved object from its silhouettes observed in perspective video sequences acquired by a Trinocular camera system.

1. Find correspondences of special, such as inflection and frontier, points on successive silhouettes of the moving object. Assume the local surface to be of second order.
2. The correspondences help obtain motion estimate/epipolar geometry.
3. The epipolar geometry is used to find correspondences elsewhere along the silhouettes. Steps 2 and 3 are iterated for refinement of the estimate.
4. An estimate of the local surface shape around a silhouetted point is also obtained.

[Larger Diagram]

Results are shown for three sequences. The figures below in each case show:

1. (left) A sample image from one of the cameras used to acquire the silhouette image.
2. (right) A video sequence showing the resulting motion and structure estimates.

(0.5MB)

(0.3MB)

(0.3MB)

1. T. Joshi, N. Ahuja and J. Ponce, "Structure and Motion Estimation from Dynamic Silhouettes under Perspective Projection", In International Journal of Computer Vision, 31(1), 1999, 31-50.