• Calendar

March 21, Tuesday
12:00 – 14:00

Multi-camera systems are an emerging class of systems with unique features. The correspondence problem is a fundamental component in solutions to many problems in multi-camera systems including calibration, object recognition, 3D scene reconstruction, and tracking. Existing vision applications in a multi-camera setting are typically based on a central computer that gathers the information from all cameras, and performs the necessary computations. Such use of a central server has a number of distinct disadvantages. One is the problem of scalability of the computational task as the number of cameras grows. Another is that the server can become a communication hot-spot and possible bottleneck. Finally, the central server is a single point of failure. The disadvantages of the central server motivate distributed solutions that are not based on a central server. In this case, reasonably powerful processors that are attached to the cameras communicate among themselves and perform all necessary computations. This scenario is quite realistic, which further motivates the distributed approach.

In this talk I will present a probabilistic algorithm for finding correspondences across multiple images. The algorithm employs the theory of random graphs to provide an efficient probabilistic algorithm that performs Wide-based Stereo (WBS) comparisons on a small number of image pairs, and then propagates correspondence information among the cameras. The method is robust to communication and processors failures, and to failures of the WBS computation. Our method can be extended to handle other distributed tasks that involve computing equivalence relation between large sets of processors

This is a joint work with: Yoram Moses, Technion Israel, Shai Avidan, IDC Israel.