A Pipelined Reconfigurable Architecture for Visual-Based Navigation

This paper is concerned with a pipelined reconfigurable architecture for implementing image processing algorithms useful for robotic navigation. The computational complexity and the real-time application of these such algorithms suggest a custom hardware approach, but the wide quantity of different algorithms useful for different tasks suggest a software reconfigurable approach. In this way, a logic reconfigurable based architecture combines both properties: hardware speed and software flexibility. Moreover, the big quantity of data to be processed using the conventional Cartesian image processing algorithms, combined with the autonomous vehicle constrains (power consumption, size and weight) makes not possible the use of reconfigurable existent machines. The log-polar vision reduces the amount of data to be processed and simplifies the computations of several interesting algorithms, allowing a reconfigurable approach which is scalable, powerful enough and small for an autonomous robot.