PaperReading

AVP-SLAM Late Fusion: Mapping and Localization using Semantic Road Marking with Centimeter-level Accuracy in Indoor Parking Lots

August 2020

tl;dr: SLAM in parking lot with road markings.

Overall impression

This work is almost exactly the same similar to AVP-SLAM. Both use road marking for mapping and localization in parking lot. Both processes 4 synchronized (or almost synchronized) fisheye images into a stitched surround view.

However it is different from AVP-SLAM in that it proposes to do semantic segmentation in the fisheye images first, perform IPM, and then do motion compensation to stitch the semantic point cloud.

Key ideas

Technical details

Notes