Parallel Algorithms for Motion Panorama Construction

A motion panorama is an efficient and compact representation of the underlying video. However, the motion panorama construction process is computationally intensive and hence extremely time consuming. Addressing this issue is crucial when one considers using motion panoramas in a real-time environment such as live video transmission. We present two parallel algorithms for motion panorama construction, namely, the shared memory parallel algorithm (SMPA) that uses POSIX threads and the distributed memory parallel algorithm (DMPA) that uses MPI. The parallel algorithms are tested on real videos. Experimental results show that the SMPA achieves linear speedup in most cases whereas the DMPA suffers from reduced efficiency when the number of processors exceeds 8.