Fast Signature Segmentation on a Multi-DSP Architecture

This paper describes the use of an embedded multiprocessor to speed up the computer vision tasks of an autonomous mobile robot. A brief overview over the robot’s vision system is given, identifying a signature segmentation algorithm as the most runtime-consuming task. A parallelization strategy for the segmentation algorithm is presented, based on the exploitation of data-parallelism. The parallel algorithm is implemented on a multi-DSP system and experimentally evaluated. Speed-up and efficiency figures are discussed.