Abstract . We present a framework for reliably androbustly detecting changes in terrain (or other 3--D objects) over time. We first present ourframework, which consists of a method formodeling terrain using a novel image-matchingmeasure called the coding loss , a method forestimating the accuracy of the resulting terrainmodels called self-consistency, and method fordetecting changes based on this estimate. Wethen present experiments using our framework. 1 Introduction The primary objective of this project is todevelop and implement a method to model anddetect changes in the shape and/or materialproperties of terrain over time.The basic approach we have adopted is to firstmodel the shape and material properties of agiven area of terrain, at given points in time,using an augmented version of our deformablemeshes (Fua and Leclerc 1994; Fua and Leclerc1995; Fua and Leclerc 1996). These modelsinclude a novel measure of the expectedaccuracy of their parameters called the self-consistency distribution
[1]
Pascal Fua,et al.
Imposing Hard Constraints on Deformable Models through Optimization in Orthogonal Subspaces
,
1997,
Comput. Vis. Image Underst..
[2]
Pascal Fua,et al.
Self-Consistency: A Novel Approach to Characterizing the Accuracy and Reliability of Point Correspon
,
1998,
ICCV 1999.
[3]
Yvan G. Leclerc,et al.
Constructing simple stable descriptions for image partitioning
,
1989,
International Journal of Computer Vision.
[4]
Pascal Fua,et al.
Taking Advantage of Image-Based and Geometry-Based Constraints to Recover 3-D Surfaces
,
1996,
Comput. Vis. Image Underst..
[5]
Pascal Fua,et al.
Registration without correspondences
,
1994,
1994 Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.