Representation And Measurement Of Non-rigid Egocentric Motion: A Parallel Implementation

Computer vision systems are by nature highly computationally intensive and hence their implementation on a parallel architecture is an attractive proposition. In addition many computer vision applications require the use of different types of parallelism (from SIMD to MIMD) at various stages of their processing. In this paper we will examine one such application, that of the analysis, representation and measurement of non-rigid motion in image sequences. In particular we examine how this application can be mapped on to U parallel architecture utilising the most appropriate form of parallelism at each stage.