GBS: Guidance by Semantics-Using High-Level Visual Inference to Improve Vision-Based Mobile Robot Localization.

摘要

The overall objective in guidance by semantics is to improve the sensing and actuation of unmanned and optionally manned platforms by incorporating high-level visual inference into the robot loop. Our specific goal in this project is to perform mapping and localization on a mobile platform using semantically meaningful sensor data. In our case, we have used a camera image co-registered with a laser scan to filter in scan points that fall on buildings in the scene or an RGBD sensor. This enables the robotic platform to only make use of readings that are known to be good, static landmarks, such as buildings as we have done in previous years. However, buildings are just one small type of semantic inference we can and plan to use.