An evaluation of ROS-compatible stereo visual SLAM methods on a nVidia Jetson TX2

摘要

Exploration vehicles need accurate localization for performing tasks such as autonomous navigation. Images captured from stereo cameras allow to estimate both the robot's motion and the environment structure. This task is referred to as Visual Simultaneous Localization and Mapping (SLAM). With this work we evaluate some of the most famous solutions for Visual SLAM compatible with the ROS (Robot Operating System) middleware. The main contributions of this work to current literature are twofold: first we evaluate the algorithm performances respectively to a nVidia Jetson TX2 board instead of a common workstation-grade computer. Secondly, we delegate image processing tasks to the embedded GPU showing significant performance gains in terms of robustness, resource utilization and processed frames per second. (C) 2019 Elsevier Ltd. All rights reserved.