Learning-Deep-Learning

AVP-SLAM: Semantic Visual Mapping and Localization for Autonomous Vehicles in the Parking Lot

August 2020

tl;dr: Semantic feature on the ground for mapping and localization in parking lots. Similar idea to Road SLAM.

Overall impression

This paper is on localization and mapping in parking lot, but the same principles apply to urban environment, which are also narrow, crowded and GPS-denied.

The semantic features are robust to perspective or illumination changes, long-term stable (as compared to traditional features such as ORB in ORB SLAM). This helps achieve cm-level accuracy required for AD.

This paper is extremely similar to a similar one from SJTU, AVP-SLAM-late-fusion. AVP SLAM requries synchronized image feeds, and AVP-SLAM-late-fusion explicitly handles unsynchronized images through late fusion of semantic point clouds.

The paper is very well written and easy to follow.

Key ideas

Technical details

Notes