Optimization of SIMD architectures for robot vision

Abstract

Optimization in Rbbot Vision Architectures is a very complex issue. For one thing we have to consider cost and performance functions. As optimization criteria we propose to work with either constant cost or constant performance. In order to obtain some interesting results we need to work with simplified, well-aproximated models of the cost and performance functions and with these models w e propose a method to assess the benefit of variations in the configuration of the processors in an SIMD array based on the trade-off between the number and the complexity of the processors. Finally, this method is applied to some aspects of the design such as the datapath width and the reconfiguration capabilities.