Real-time implementations of an MRF-based motion detection algorithm

The main concern in image processing is the computation cost. Markov random field (MRF)-based algorithms particularly require a significant amount of computation. The paper investigates three solutions to implement a simple, but robust, MRF-based motion detection algorithm in real time: SIMD machine, DSP-based image processing board, and analog resistive network. Details and performances of each implementation are given and a comparison between each realization is made. The underlying goal of this work is to study if real-time implementations of MRF-based algorithms are feasible or not. The answer is positive in the case of quite simple algorithms, but reserved with more complex ones. (C) 1998 Academic Press Limited.