Multi-Body Motion Estimation and Segmentation from Multiple Central Panoramic Views

Omid Shakernia and Rene Vidal
(Professor S. Shankar Sastry)
(ONR) N00014-00-1-0621

In [1], we present an algorithm for infinitesimal motion estimation from multiple central panoramic views. We first derive the optical flow equations for central panoramic cameras as a function of both pixel coordinates and back-projection rays. We then derive a rank constraint on the optical flows across many frames, which must lie in a six dimensional subspace of a higher-dimensional space. We then propose a factorization approach for recovering camera motion and scene structure.

In [2], we generalize this approach to the case where there are multiple independently moving objects. We show that one can apply a factorization technique to estimate the number of independently moving objects and segment the omnidirectional optical flow based on the fact that the flows generated by independently moving objects lie in orthogonal subspaces to each other.

We present experimental results of the proposed motion segmentation and estimation algorithm on a real image sequence with two independently moving mobile robots, and evaluate the performance of the algorithm by comparing the vision estimates with ground-truth GPS measurements gathered by the mobile robots.


Figure 1: Showing the optical flow of an omnidirectional camera viewing two independently moving mobile robots

Figure 2: Showing the multi-body motion segmentation of the mobile robots

[1]
O. Shakernia, R. Vidal, and S. Sastry, "Infinitesimal Motion Estimation from Multiple Central Panoramic Views," IEEE Workshop on Vision and Motion Computing, Orlando, FL, December 2002.
[2]
O. Shakernia, R. Vidal, and S. Sastry, "Multi-Body Motion Estimation and Segmentation from Multiple Central Panoramic Views," IEEE Conf. Robotics and Automation, 2003 (submitted).

More information (http://robotics.eecs.berkeley.edu/~omids) or

Send mail to the author : (omids@eecs.berkeley.edu)


Edit this abstract