A reconfigurable and hierarchical parallel processing architecture: performance results for stereo vision
 Proceedings. 10th International Conference on Pattern Recognition
The degree of exploitable parallelism in computer vision tasks varies with time and image position. Therefore, an architecture for vision must be highly flexible and modular. In this paper we consider one such multiprocessor architecture (NETRA) which is highly reconfigurable and does not involve the use of complex interconnection schemes. The topology of this multiprocessor is recursively defined, and hence, is easily scalable from small to large systems. It has a tree-type hierarchical
... hierarchical architecture each of whose leaf nodes consists of a cluster of small but powerful processors connected via programmable crossbar with selective broadcast capability. The architecture is simulated on a hypercube multiprocessor and the performance of one processor cluster is evaluated for stereo vision tasks. The particular stereo algorithm selected for implementation requires computation of two-dimensional Fast Fourier Transform (2D-FFT), template matching, histogram computation and leastsquares surface fitting. We use static partitioning of data for the data independent tasks such as 2D-FFT and dynamic scheduling and load balancing for the data dependent tasks of feature matching and disambiguation. The superior performance of our architecture is demonstrated by comparing the results with that of a similar implementation on the hypercube itself.