3D Nonrigid Motion Analysis from 2D Images

Lin Zhou and Chandra Kambhamettu

Collaborator: Dmitry B. Goldgof

Related Publications

The accurate calculation of structure and nonrigid motion from a sequence of images is a common problem in computer vision and shows up in many real-world applications. This work focuses on extracting this motion and structure from 2D image sequences without known correspondences, with a particular focus on cloud sequences from satellite imagery.

In order to retrieve the desired structure and nonrigid motion, we break the analysis into a global module to constrain the non-rigid behavior and a local module to extract the fine details. The estimated structure and motion in the images is obtained via an iterative algorithm which fuses the two modules into a global-local framework. Our algorithm is implemented in a structure and motion analysis system (SMAS) that is used to perform nonrigid motion and structure analysis on sequences of cloud motion.

Fig. 1 on the left shows one of a sequence of images obtained at one-minute intervals from the satellite GOES-9 during Hurricane Luis on September 6, 1995. Fig. 2 on the right shows the recovered structure of the image using our SMAS system.