Performance of optical flow techniques
International Journal of Computer Vision
PVM: Parallel virtual machine: a users' guide and tutorial for networked parallel computing
PVM: Parallel virtual machine: a users' guide and tutorial for networked parallel computing
Multiple Constraints to Compute Optical Flow
IEEE Transactions on Pattern Analysis and Machine Intelligence
Motion and Structure from Image Sequences
Motion and Structure from Image Sequences
A Practical Approach to Dynamic Load Balancing
IEEE Transactions on Parallel and Distributed Systems
Robust Discontinuity-Preserving Model for Estimating Optical Flow
ICPR '96 Proceedings of the 1996 International Conference on Pattern Recognition (ICPR '96) Volume I - Volume 7270
| Hi-index | 0.00 |
Motion estimation for image sequences has several important applications (vision, tracking, 3D reconstruction, indexing...).When high quality motion vector fields are required for large images, the computation load is huge: parallelism gives results in acceptable time. Which architecture to use depends on time constraints, architecture suitability to the algorithm, etc. This paper reports current work on parallelizing a multi-resolution, multi-attribute method, on two different architectures, one a massively parallel computer, the other a network of workstations. Although of older a technology, the parallel computer has various communication possibilities and a large bandwidth. On the other hand, networks of workstations are a possibly emerging standard for lowcost super-computing. Comparison of both shows that, for this kind of algorithm, even with some load balancing they provide less gain if communicating only via a classical network. The next stage of this work is to address high bandwidth networks.