We present a geometry-based indexing approach for the retrieval of video da
tabases. It consists of two modules: 3D object shape inferencing from video
data and geometric modeling from the reconstructed shape structure. A moti
on-based segmentation algorithm employing feature block tracking and princi
pal component split is used for multi-moving-object motion classification a
nd segmentation. After segmentation, feature blocks from each individual ob
ject are used to reconstruct its motion and structure through a factorizati
on method. The estimated shape structure and motion parameters are used to
generate the implicit polynomial model for the object. The video data is re
trieved using the geometric structure of objects and their spatial relation
ship. We generalize the 2D string to 3D to compactly encode the spatial rel
ationship of objects. (C) 2000 Academic Press.