Path Planner for Keyframe-based Visual Autonomous Navigation
Min Gyung Jang, Hee-Won Chae, Jae-Bok Song · 2019
Visual navigation systems have received much attention in recent years. Such systems generate keyframes storing the sensor information that provides a way to correct the robot pose. However, the conventional gradient path does not generate a path that tracks the keyframes since it is based only on the costs regarding the environment. In this study, we propose a keyframe vector-based path planner (KVPP) that is more suitable for visual navigation systems as this path follows more keyframes in the map to increase the chance of keyframe-based pose correction during autonomous driving. This KVPP path uses the existing keyframes as a reference to path generation. Various experiments were conducted to evaluate the KVPP in the real environment and were compared with the conventional gradient path to verify its effectiveness.