ABSTRACT
Real-time computer vision applications are characterized by a rapid response to information derived from a high-speed continuous data stream from a sensor. The associative string processor (ASP) modules comprise highly versatile parallel processing building blocks for the simple construction of fault-tolerant, massively parallel processors capable of TeraOPS performance and support of continuous data input. The requirements for rapid response to visual information derived from high-speed continuous data streams and the performance limitations of general-purpose uniprocessors led to parallel processing architectures for computer vision applications. The ASP implementation of the zero-crossing detection is very much simplified using the associative property of the architecture. An ASP module comprises a multiple-instruction control of multiple single instruction multiple data nodes and a parallel processing structure of intercommunicating ASP substrings, each supported with an ASP data buffer and an ASP control unit. The ASP implementation enhances the performance of the convolution algorithm using scalar-vector computation to calculate the required sum.
