Load balancing algorithm of parallel vision processing system for real-time navigation

摘要

We present a load balancing algorithm called perpendicular image partitioning (PIP) for parallel vision processing. Image partitioning plays two roles: load distribution and load balancing. Image partitioning is executed not simply by the number of pixels, but by the number of interested pixels, or features. The characteristics of load variance are analyzed, and the positions of two perpendicular (vertical, horizontal) lines are determined using these characteristics. The PIP algorithm offers fast load balancing, and enables efficient vision processing with well balanced load. We apply the PIP algorithm to real-time depth determination for mobile robot navigation with a sequence of images.