Communication in a multi-layer MIMD system for computer vision

Algorithms in computer vision are computationally intensive and have distinct characteristics. Message-passing MIMD computers are popular architectures for computer vision due to their flexibility and asynchronous structure. However, performance of parallel programs executed on these systems is dependent on the efficiency of inter-processor communication. This paper describes a software system developed to handle complex communication mechanisms for a system designed for computer vision applications. The system is constructed with TI320C40 DSPs and employs the distributed memory MIMD paradigm. A set of experiments has been conducted to determine the hardware parameters of the communication links of the processors. Furthermore, the performances of the communication algorithms developed, which are frequently required in parallel vision algorithms, are presented. These results will be useful in optimizing performance of parallel programs developed for the system.