Efficient Stereoscopic Video Matching and Map Reconstruction for a Wheeled Mobile Robot

摘要

This paper presents a novel method to achieve stereoscopic vision for mobile robot (MR) navigation with the advantage of not needing camera calibration for depth (distance) estimation measurements. It uses the concept of the adaptive candidate matching window for stereoscopic correspondence for block matching, resulting in improvements in efficiency and accuracy. An average of 40% of time reduction in the calculation process is obtained. All the algorithms for navigation, including the stereoscopic vision module, were implemented using an original computer architecture for the Virtex 5 FPGA, where a distributed multicore processor system was embedded and coordinated using the Message Passing Interface.